[0:00] hi welcome back to the channel my name
[0:01] is Bogdan and in this video I'll provide
[0:04] you with the most comprehensive piece of
[0:06] content on how to master AI chatbots if
[0:09] you watch this video from start to
[0:11] finish you'll be able to clearly
[0:14] understand how AI chatbots work how to
[0:16] build them and which kinds of chatbots
[0:18] you can sell to businesses there is a
[0:20] great tutorial on AI chatbots published
[0:22] by leam Motley it covers a lot of Basics
[0:25] so if you haven't checked it out yet
[0:27] feel free to do so but it was published
[0:29] almost a year year ago and since then
[0:31] we've had a lot of updates Liam and I
[0:34] actually discussed this recently how
[0:36] rapid the technology advancement is I
[0:38] mean in the last 10 months we've seen
[0:40] more capable GPT models higher token
[0:42] limits new tools Vision capabilities
[0:44] once GPT 40 is fully released will have
[0:47] audio input and output and all of these
[0:50] updates obviously create new use cases
[0:52] for businesses so long story short this
[0:55] video is going to be an updated version
[0:58] of a full guide on AI chatbots my team
[1:01] and I worked on this video for over a
[1:04] month we could have made it a paid
[1:06] course but we decided to offer it for
[1:08] free well not exactly for free you pay
[1:11] for it you pay for it with your
[1:12] attention which is the currency of today
[1:14] right so make sure your investment pays
[1:17] off watch this video as many times as
[1:20] you need but understand the information
[1:22] here and act on it remember knowledge
[1:25] alone doesn't equal results your actions
[1:28] do and for full transparency my business
[1:30] interest here is clear I've been
[1:32] building IT solutions for The Last 5
[1:34] Years with my current CTO we build a SAS
[1:37] product a Marketplace and now we run an
[1:40] AI automation agency you can check it
[1:42] out at boss. agency what we essentially
[1:45] do is build these AI chat Bots voice
[1:48] Bots and automation Solutions and we
[1:51] sell them to various businesses we have
[1:53] a team of developers we are fully
[1:55] equipped to take on more complex
[1:57] Solutions and we are not me specific so
[2:01] the more leads we get the better if you
[2:03] watch this video and find it valuable
[2:05] YouTube algorithm will push it more I'll
[2:08] get more attention and more leads for my
[2:10] business I hope that makes sense to you
[2:12] and explains why we spend a ridiculous
[2:15] amount of time and effort to prepare
[2:17] this comprehensive guide so what we'll
[2:20] cover today first we'll discuss the AI
[2:23] business potential if you clicks on this
[2:24] video you already realize the potential
[2:27] I know so I'll just quickly share some
[2:29] important stuff
[2:30] and explain what it means for you if you
[2:32] want to leverage this AI opportunity
[2:35] then we'll dive into understanding AI
[2:37] chatbot this will be the theoretical
[2:40] part of the video but I will only
[2:42] discuss the parts that you'll deal with
[2:44] in practice okay you absolutely must
[2:47] understand and be able to talk about
[2:49] these aspects because in 90% of cases
[2:52] you'll touch on them during sales calls
[2:55] of course if your goal is to build and
[2:57] then sell these shots to businesses if
[2:59] you if you want to learn more about llms
[3:01] promting neural networks and other
[3:04] fundamentals check out the suggested
[3:06] videos in the description I highly
[3:08] recommend Andre kpa's 1hour talk on
[3:11] intro to large language models also
[3:13] check out the IBM Channel they have a
[3:15] great playlist called understanding AI
[3:18] models and one of my favorite channels
[3:20] is three blue one brown which also has a
[3:23] fantastic playlist on neural networks I
[3:25] also included a chapter on prompt
[3:27] engineering because it is a skill it is
[3:29] a skill that you need to learn in order
[3:31] to build efficient AI chatbots I'm going
[3:34] to share some tips and hacks on
[3:36] prompting that can save you a lot of
[3:37] money so make sure to use them next I'll
[3:40] share some interesting use cases all of
[3:43] them are from real life experiences as
[3:46] we get daily leads looking to implement
[3:48] AI Solutions in their businesses we have
[3:51] plenty of stories to share and then in
[3:53] the Practical part of the video I'll
[3:55] review the most useful tools for
[3:57] building and deploying AI chbs I'll show
[4:00] you how these tools have evolved over
[4:02] the past year adding some very
[4:04] interesting features finally we get to
[4:06] the most exciting part the Practical
[4:09] tutorials you might be tempted to skip
[4:12] everything and just jump right to this
[4:14] but I urge you to give the theory part a
[4:16] chance because you need to understand it
[4:19] before implementing these solutions for
[4:21] the tutorials I'll guide you through
[4:23] comprehensive Solutions step by step
[4:26] sharing my screen first I'll use a no
[4:29] code chatbot Builder to create a basic
[4:31] customer service chatbot then I'll show
[4:33] you how to build a more advanced chatbot
[4:36] introducing a product recommendation
[4:38] algorithm after that I'll demonstrate
[4:41] how to use custom code and different
[4:43] llms such as Cloe and Gemini explaining
[4:47] which cases each llm is best suited for
[4:51] and eventually I'll show you a
[4:52] comprehensive AI chatboard capable of
[4:55] providing customer support recognizing
[4:57] images and recommending real products
[5:00] based on the customers needs so we've
[5:03] packed a few tutorials into one
[5:05] comprehensive video to make it as
[5:07] valuable for you as
[5:10] possible I could spend a lot of time
[5:12] discussing AI Trends and growth rates
[5:15] and by the way the AI Market is
[5:17] estimated to hit $1 trillion in the next
[5:19] six years numerous studies show that the
[5:22] adoption rate of AI am businesses will
[5:25] keep increasing as shown in this slide
[5:27] but I will only say one word that should
[5:30] matter most timing obviously this is a
[5:33] technology Revolution every company is
[5:36] diving into it we just had Apple's WWDC
[5:39] event introducing Apple intelligence
[5:42] which means hundreds of millions of
[5:44] people will start using AI accelerating
[5:47] Mass adoption even faster it's clear
[5:49] that we are quickly heading into a world
[5:51] where you can just take your phone talk
[5:53] to it and it responds intelligently
[5:56] knowing you and historically the
[5:58] penetration of new technology ology
[6:00] typically begins with businesses to
[6:01] Consumer b2c Market before expanding
[6:04] into the B2B sector businesses to
[6:06] business it happened with social media
[6:09] platforms it happened with the internet
[6:11] and it's just beginning to happen with
[6:13] AI in the last 20 years we've seen
[6:16] businesses go digital companies made
[6:18] websites set up social media profiles
[6:20] and moved from offline to online ads the
[6:24] next big thing is going to be automation
[6:27] especially with AI getting into to
[6:29] business processes this change aims to
[6:32] cut costs save time and overall boost
[6:35] efficiency and if we look back at the
[6:37] early days of going digital companies
[6:40] that focus on making websites and doing
[6:42] social media marketing made a lot of
[6:45] money because the timing was right now
[6:47] we have a similar opportunity with AI
[6:49] agencies that focus on AI and automation
[6:52] will be key in helping businesses get on
[6:54] board with these new technologies as
[6:56] more businesses start using AI you want
[6:58] to get in as early as possible this
[7:01] chart shows the technology adoption
[7:03] cycle and we are still at the early
[7:05] adoption stage which represents only
[7:08] 13.5% of the market when the early
[7:10] majority and then the late majority
[7:12] start looking to implement AI Solutions
[7:14] you want to already be established as an
[7:17] expert with track record with case
[7:20] studies in your portfolio again right
[7:22] now only a small number of businesses
[7:24] use AI but that's changing fast so the
[7:28] best time to start building and selling
[7:30] these Solutions particularly AI chatbots
[7:33] is
[7:35] now all right let's talk about chatbot
[7:37] just to make sure we're on the same page
[7:40] we've got two types here old school
[7:42] rule-based and the new AI powered ones
[7:45] the old school rule-based chwat are
[7:47] pretty limited and manual they work by
[7:49] following a set of predefined rules
[7:51] which means they can't handle anything
[7:54] outside those rules on the other hand AI
[7:56] powered chatbots use large language
[7:59] model to understand the user's query and
[8:02] provide an answer that's obvious right
[8:04] basically chat GPT is also a chatboard
[8:06] but for our Solutions we connect the
[8:09] same GPT model to a different interface
[8:11] with a different context now I'm going
[8:13] to simplify a lot of things but I'll try
[8:15] to structure how it all works so that
[8:17] you understand the main logic I like to
[8:19] think of it as three main elements user
[8:22] prompt knowledge based and llm so it
[8:24] works like this the user asks a question
[8:27] or makes a request then the chat
[8:29] searches its knowledge base to
[8:31] understand the prompt next the AI
[8:34] processes The Prompt and uses the
[8:35] knowledge base to create an answer and
[8:38] finally the chatboard provides the
[8:39] answer to the user but the issue is the
[8:42] token limitations every llm has a token
[8:45] limit you probably heard of it for
[8:46] example GPD 3.5 had a limit of 4,096
[8:50] tokens the latest and most advanced GPT
[8:53] 40 has a
[8:56] 128,000 token limit so compared to a
[8:59] year ago you've got much more
[9:01] flexibility when it comes to token usage
[9:04] but keep in mind tokens get used up for
[9:07] three things the first one is processing
[9:09] the user's query the longer you prompt
[9:11] the more tokens will be used second
[9:13] pulling information from your knowledge
[9:15] base tokens are used both for quering
[9:18] the knowledge base and for the
[9:19] information it retrieves and third
[9:22] generating a response so the length and
[9:25] complexity of the response consume
[9:27] tokens too including interpreting your
[9:29] input and adding relevant info from your
[9:32] knowledge base and even though GPT 40 is
[9:35] twice as cheap as the previous gp4 turbo
[9:38] it's still 10 times more expensive than
[9:41] GPT 3.5 turbo so if you can achieve your
[9:44] goals using a cheaper model it's always
[9:46] better to go for it to get around the
[9:48] token limits we use chunking it means
[9:51] that your knowledge base is split into
[9:53] chunks of text and the AI picks only the
[9:57] relevant chunks to answer the the users
[10:00] prompt that's why our still high level
[10:03] but slightly more detailed framework for
[10:06] the chatbot looks like this the first
[10:08] step the user enters a prompt the Second
[10:12] Step break the knowledge base into
[10:14] smaller chunks the next step is to
[10:16] retrieve the most relevant chunks based
[10:19] on the users's prompt then create a new
[10:21] prompt that includes the users's
[10:23] question and the relevant context from
[10:26] our knowledge base then feed the new
[10:29] promp to the language model and finally
[10:31] return the generated answer to the user
[10:33] let's visualize it here so we have
[10:35] knowledge base plus user prompt then the
[10:38] system creates a context aware prompt So
[10:41] based on the relevant chunks of our
[10:44] database to the user prompt and then llm
[10:47] generates the final result but the
[10:50] problem is with step number two how do
[10:52] we decide which chunk of text is
[10:55] relevant to the user's query a common
[10:57] solution is to use embeddings embeddings
[11:00] capture the semantical aspects of texts
[11:03] let's use this graph to explain each day
[11:05] of the week is represented as a point in
[11:07] space okay the positioning of these
[11:10] points shows how closely related their
[11:12] meanings are for instance the days
[11:15] Monday Tuesday Wednesday Thursday and
[11:17] Friday are grouped closely together
[11:20] indicating that they are semantically
[11:23] similar they are all weekdays right
[11:26] similarly Saturday and Sunday are also
[11:29] close to each other representing the
[11:31] weekend and they are all close together
[11:33] because they are all days of the week
[11:36] the embeddings work by converting words
[11:38] phrases or other pieces of text into
[11:41] numerical vectors as you can see on the
[11:44] right hand side here these vectors
[11:46] capture the semantic meaning of the text
[11:49] and when displayed in a
[11:51] multi-dimensional space like in this
[11:53] slide similar meanings are placed near
[11:56] each other so let's summarize just to
[11:58] make sure we are in the the same page
[12:00] words or phrases that have similar
[12:02] meanings or context will be positioned
[12:05] close to each other the further apart
[12:07] two points are the less similar their
[12:10] meanings are and all the words or
[12:12] phrases are given an embedding Vector
[12:15] the numerical one right and by comparing
[12:17] the distance between two embedding
[12:19] vectors you can measure how similar
[12:21] their meanings are I hope that's clear
[12:24] it is important to understand this if
[12:25] you actually want to solell these
[12:27] Solutions in real life okay now we have
[12:29] an updated highlevel framework for our
[12:31] chatbot when the user enters a prompt
[12:34] the the system starts by chunking right
[12:36] which means it divides large texts like
[12:38] it breaks your large knowledge base into
[12:41] smaller manageable pieces then it
[12:43] converts the data into numerical vectors
[12:46] or embeddings that capture their
[12:48] meanings and this allows the system to
[12:50] compare and retrieve similar information
[12:53] effectively next the system creates an
[12:55] embedding for the user prompt and
[12:57] searches the embedding database for the
[12:59] chunks of information that are closest
[13:02] to this prompt embedding it retrieves
[13:04] the actual text of the most relevant
[13:06] chunks and creates a new prompt that
[13:08] combines the users's question with
[13:11] context from the database this revised
[13:13] prompt is then sent to the language
[13:15] model which generates the answer so the
[13:18] key is not that the AI knows everything
[13:21] but that it's smartly retrieves and uses
[13:24] the most relevant information it's like
[13:26] a librarian fetching the right books for
[13:28] you rather than knowing everything off
[13:31] the top of their head all right guys if
[13:32] you managed to understand this you can
[13:34] be proud of yourself there are a lot of
[13:36] people talking about AI on YouTube who
[13:39] don't even get this high level framework
[13:41] moving on let's discuss
[13:44] prompting I promise to only touch on the
[13:47] theory you need to actually build these
[13:49] chatbots in practice and effective
[13:52] prompting is one of those essential
[13:54] things because it directly impacts the
[13:57] cost and the efficiency of your chat
[13:59] Bots I used to think it was
[14:01] straightforward there are so many videos
[14:04] with perfect formulas for promps on
[14:07] YouTube and you know if your goal is to
[14:09] just use CH GPT to revise your emails
[14:12] that might be enough but if you want to
[14:14] build AI assistants put them to work or
[14:17] even sell them to clients you need to
[14:19] understand that there is a science
[14:21] behind it and if you don't learn it
[14:23] you'll struggle to get the proper cost
[14:25] efficiency ratio to be able to sell to
[14:27] anyone so I want to break it down for
[14:29] you and while doing that we will write a
[14:32] good prompt which we will use later in
[14:34] this video when I build a chatbot live
[14:36] the chatbot will serve as an online
[14:38] Beauty Store consultant for an imaginary
[14:40] brand bosar cosmetics it will be able to
[14:44] provide customer support recognize
[14:46] images and recommend relevant products
[14:48] to users what I'm sharing with you now
[14:51] is based on research papers not just my
[14:53] own experience scientists have tested
[14:56] various prompting techniques and
[14:58] measured their impact on efficiency in
[15:00] the video description I'll provide links
[15:02] to all these research papers so you can
[15:05] check them out yourself there are two
[15:06] types of prompt engineering
[15:08] conversational and single shot
[15:10] conversational prompting is suitable for
[15:12] small tasks or personal use you know
[15:15] when when you ask CH GPT to fine tune
[15:17] your email and if it doesn't do well on
[15:20] the first try you can follow up until
[15:22] you get the output you want single shot
[15:24] prompting however is important for
[15:26] automating systems and creating scalable
[15:29] AI Solutions this is what we aim to do
[15:31] here this method involves crafting a
[15:34] prompt that provides all the necessary
[15:36] information in one go which is essential
[15:39] for large scale and kind of more complex
[15:42] applications there are no follow-ups
[15:45] okay so the main components of a good
[15:47] prompt are role task specifics context
[15:51] examples and noes each of these
[15:54] components is supported by a prompt
[15:56] technique that has been researched and
[15:58] backed by scientific papers these
[16:01] techniques are rooll prompting Chain of
[16:04] Thought prompting emotion prompt F short
[16:07] prompting and loss in the middle effect
[16:09] let's quickly cover each of these
[16:11] components and the relevant prompt
[16:13] technique and then we'll move on to the
[16:14] next chapter the first component is roll
[16:17] and the relevant prompt technique is
[16:18] roll prompting roll prompting is a
[16:21] technique where the language model is
[16:23] assigned a specific role to play during
[16:25] the interaction for example you are a
[16:28] high highly qualified and experienced
[16:30] online Beauty Store consultant you are
[16:33] the best at selecting the perfect Beauty
[16:35] and makeup products to meet each
[16:37] customer's unique needs super simple I
[16:39] know you are already familiar with this
[16:41] technique but make sure to use it
[16:43] because research shows it can increase
[16:45] output accuracy by up to
[16:48] 25% this is especially true if you not
[16:50] only describe the role but also provide
[16:53] a complimentary description of their
[16:55] abilities complimentary description so
[16:58] in my case the role is you are a highly
[17:00] qualified and experienced online Beauty
[17:03] Store consultant and the complimentary
[17:05] description is you are the best at
[17:07] selecting the perfect Beauty and makeup
[17:09] products to meet each customer's unique
[17:12] needs the next component is Task and the
[17:14] correlating prompt technique is Chain of
[17:16] Thought prompting that's where we tell
[17:19] it what to do Provide support generate
[17:21] text Etc it should be concise and
[17:24] specific okay Chain of Thought prompting
[17:27] involves instructing the model to think
[17:30] step by step essentially giving it a
[17:32] detailed process to follow research
[17:34] shows this technique can boost output
[17:36] accuracy by up to
[17:38] 90% for complex tasks which is a
[17:41] significant boost right here's a
[17:43] screenshot from one of the research
[17:44] papers I mentioned it shows an example
[17:47] of this prompting technique where Chain
[17:49] of Thought reasoning is highlighted so
[17:51] we can see the difference in handling an
[17:53] arithmetic task but let's have a look at
[17:55] our example okay so I always start with
[17:58] a verb provide customer service and
[18:00] advice on services available at bossar
[18:02] cosmetics and then I provide a
[18:05] step-by-step process instruction follow
[18:08] this step-by-step process to ensure your
[18:10] script is first class first greet the
[18:13] customer warmly and answer any questions
[18:16] they might have step two identify
[18:18] customers needs ask what kind of beauty
[18:21] products they are looking for skin care
[18:23] makeup hair care or something else step
[18:26] three gather detailed information ask
[18:28] them about their skin type specific
[18:30] concerns and the look they are aiming
[18:33] for step four request an image of their
[18:36] face for better assessment because we
[18:38] are going to recognize images Step five
[18:40] suggest products based on the customer's
[18:42] needs and available products in the
[18:44] store step six explain how the
[18:47] recommended Products address their
[18:49] specific concerns or solve their pain
[18:51] points step seven let them know they can
[18:53] reach out for further assistance after
[18:56] their purchase next we have specifics
[18:58] and the associated prompt technique is
[19:01] emotion prompt this section is where you
[19:03] can list the most important notes about
[19:05] executing the task outlined above in our
[19:08] case we can add specifics such as check
[19:10] the product database before recommending
[19:12] products to ensure they are in stock or
[19:15] if you can't find the right products to
[19:17] satisfy the customer needs encourage
[19:19] them to search the site themselves the
[19:22] emotion prom technique involves adding
[19:24] short phrases or sentences with
[19:26] emotional stimuli to the original prompt
[19:29] this method has been shown to boost the
[19:32] accuracy of generated output by up to
[19:35] 115% for complex tasks so you can get
[19:38] better results by adding more bullet
[19:40] points with phrases like your role is
[19:43] vital for the whole company both I and
[19:46] our customers greatly value your
[19:49] assistance and recommendations I know it
[19:51] sounds strange and you might think it's
[19:54] nonsense but I encourage you to check
[19:56] out the research papers to see how was
[19:58] measured and studied for the next
[20:00] section context we're going to combine
[20:03] both rooll prompting technique and
[20:05] emotion prompt this section's goal is to
[20:08] give context about the environment our
[20:10] llm is working in and why it's doing its
[20:14] specific task we want to explain its
[20:17] role within our business context and
[20:19] kind of hype it up adding some
[20:22] additional stimul to show how important
[20:24] the chatbot role is so we want to use
[20:26] phrases like you are the world class
[20:29] assistant and your expertise is highly
[20:31] important to the company or you are the
[20:34] most important component of our business
[20:36] processes people that you are advised to
[20:39] rely on you as never before something
[20:42] like that here's my example our company
[20:45] sells highquality Cosmetics like skin
[20:47] care makeup Hair Care and More We value
[20:50] our customers and our goal is to solve
[20:53] their pain points that part provides
[20:56] context about my business then Your Role
[20:59] is to provide customer service
[21:01] understand customer needs and recommend
[21:03] products that meet those needs here I
[21:06] describe its role within our business by
[21:09] accurately identifying customers needs
[21:11] you directly contribute to their
[21:13] well-being and the growth and success of
[21:15] our company therefore we greatly value
[21:18] your attention to customer service and
[21:20] need identification and here I add some
[21:23] emotional stimuli to show how important
[21:25] its role is so basically there are two
[21:28] things things to remember about context
[21:30] explain the chatbots role in the
[21:31] business context including details about
[21:33] customers types of services or Products
[21:36] Company values Etc then emphasize its
[21:39] importance with emotional appeal okay so
[21:42] you want to highlight its impact on the
[21:44] business and The Wider Community or even
[21:46] the whole society if that makes sense
[21:49] next section is examples and the
[21:51] associated technique is few shot
[21:53] prompting which essentially means that
[21:55] we provide several examples while zero
[21:57] shot prompting means there are no
[21:59] examples given and one shot means there
[22:02] is one example provided so according to
[22:05] studies the accuracy can be increased by
[22:07] up to 57% if you provide multiple
[22:11] examples so on the graph here you can
[22:13] see that the accuracy can be increased
[22:15] by 40 something per if you go from zero
[22:18] examples to at least one and then if you
[22:21] add more you can get even higher
[22:23] accuracy one thing you should remember
[22:25] though is that token usage involves
[22:27] processing your prod prompt we discussed
[22:29] it already right so the more taex you
[22:31] include in your prompt the higher the
[22:33] token usage and you pay for those tokens
[22:35] so keep the prompt as brief as possible
[22:37] while implementing all these techniques
[22:39] to achieve the best result you could add
[22:41] thousands of examples but then your
[22:43] prompt will be huge and it will be more
[22:46] expensive to process it so in practice
[22:48] we use four to five examples on average
[22:50] depending on the context and usually it
[22:53] is enough to achieve the best
[22:54] performance and also it is important to
[22:56] provide examples that the system
[22:58] struggles with we usually start by
[23:00] testing it during testing we identify
[23:03] the types of queries that are most
[23:05] difficult for the model to answer then
[23:07] we take those queries and provide the
[23:09] ideal outputs as examples in the prompt
[23:12] just a little life hack for you and this
[23:14] is also an opportunity for us to teach
[23:16] it how to structure the output by
[23:19] providing specific examples for our
[23:21] beauty store we could provide examples
[23:23] such as these so it could be typical
[23:25] question by the customer and then the
[23:27] ideal of output the ideal answer by the
[23:30] chatbot so hi I have really dry skin
[23:33] especially during the winter months can
[23:35] you recommend some products to help with
[23:38] hydration and then the ideal answer by
[23:40] the chatbot I won't waste your time
[23:42] reading out loud the the rest of the
[23:44] examples you can pause and just check
[23:46] them out let's move on to the last
[23:47] section which is for nodes this is your
[23:50] final opportunity to remind the model of
[23:52] key points and add any final guidelines
[23:54] to get the output right also it's a good
[23:57] place for some cool hacks I like to
[24:00] include things like letting the model
[24:02] say I don't know this is a great way to
[24:05] prevent hallucinations you allow it to
[24:07] say I don't know instead of making
[24:09] things up so definitely use it then
[24:12] giving it room to think this allows the
[24:15] model to draft better responses because
[24:17] you kind of allow it to take time and
[24:19] think about it other words you allow it
[24:21] to use this step-by-step thinking
[24:23] process and once again be encouraging
[24:26] remember you are the world class expert
[24:29] in X that helps a lot one thing to keep
[24:32] in mind is the loss in the middle effect
[24:35] studies show that language models do
[24:37] best when important information is at
[24:39] the start or end of The Prompt if your
[24:42] prompt is long stuff in the middle might
[24:45] get overlooked as you can see on this
[24:47] graph which is again a screenshot from a
[24:49] research paper I didn't make it up so
[24:53] keep the notes section short and focus
[24:55] on the most important functions and the
[24:57] style you want okay for our example I
[25:00] could come up with notes like this if
[25:02] you don't have the answer to a query you
[25:04] can say I don't have an answer please
[25:06] send your query at support bossar
[25:09] cosmetics.com then before answering the
[25:12] query take a deep breath and think
[25:14] through it step by step okay then you
[25:18] are the world class expert in beauty
[25:21] industry and something like your tone
[25:23] should be friendly and your main goal is
[25:26] provide the best customer service
[25:28] service all right this is our final
[25:30] example prompt broken into sections
[25:32] which we will use later when building
[25:34] chatbots and finally a few more tips
[25:37] regarding prompting number one Implement
[25:39] all of the techniques we've talked about
[25:42] if you do that you can boost your
[25:44] performance by up to 300% number two
[25:47] prompt length and cost so for high
[25:49] volume tasks keep your prompt short and
[25:52] to the point because obviously each time
[25:54] it runs you are charged for the input
[25:56] tokens so a shorter prompt means lower
[25:59] costs and we always want to keep the
[26:02] system as cheap as possible as long as
[26:04] it can complete the task okay number
[26:07] three be smart about choice of model
[26:10] good prompt engineering can make cheaper
[26:12] models work better here's a strategy on
[26:15] how we sometimes approach this let's say
[26:18] we use openis models start by testing
[26:21] GPT 3.5 turbo then test GPT 40 and if
[26:25] you notice any difference if it does the
[26:27] job better then you can use the results
[26:31] from GPT 40 as examples within the
[26:35] prompt for GPT 3.5 turbo that way you
[26:39] can achieve the same results as if you
[26:41] used GPT 40 using GPT 3.5 turbo for your
[26:46] specific examples and you would save a
[26:48] ton of money because GPT 3.5 turbo is 10
[26:51] times cheaper than gp40 and then we have
[26:53] temperature be a chatboard builder or an
[26:56] open AI Dev platform while creating an
[26:59] assistant you'll often find this
[27:01] temperature in the model configuration
[27:03] temperature controls Randomness so as
[27:06] the temperature approaches zero the
[27:09] model will become deterministic and
[27:11] repetitive and that is what we usually
[27:13] want because we aim to achieve
[27:16] consistent and predictable results with
[27:18] our AI system in most cases so we
[27:21] usually set the temperature to zero one
[27:23] of the exceptions might be if you want
[27:25] to do some kind of creative wri or
[27:28] ideation or something like that okay so
[27:31] then you can test higher levels for the
[27:33] temperature but usually by default we
[27:35] set it to
[27:37] zero now that you understand how AI
[27:40] chatbots work and how to craft and test
[27:43] the best prompts let's quickly review
[27:45] the use cases there are obvious benefits
[27:48] like improved customer engagement like
[27:51] 24/7 availability right they offer
[27:54] global Service by communicating in
[27:56] multiple languages and of course you can
[27:58] save on costs by reducing the need for a
[28:00] large customer service team these shads
[28:03] can increase Revenue through
[28:05] personalized Communications and upsells
[28:08] and at the same time provide data
[28:10] analytics to you these benefits are
[28:13] already significant and they come from
[28:15] basic AI chart bards but on top of that
[28:18] you can build a lot more automation for
[28:21] example lead qualification and
[28:23] customized sales funnels are among the
[28:25] most popular requests we get at our
[28:27] agent gen to convert leads effectively
[28:29] businesses need to Target them with
[28:31] tailored sales funnel right to address
[28:34] their specific pain points a general
[28:36] sales funnel for All Leads results in
[28:39] lower conversions for instance at our
[28:42] agency we build chatbots for customer
[28:44] support we build voice Bots to handle
[28:47] calls and serve as receptionists and we
[28:50] also automate social media management
[28:53] different leads are interested in
[28:54] different solutions an AI chatboard can
[28:58] qualify leads identify their specific
[29:01] needs and run them through customized
[29:03] sales funnels offering targeted
[29:05] Solutions instead of a one- siiz fits
[29:08] all approach and that is how you can
[29:10] help businesses dramatically increase
[29:12] their conversion rate another great
[29:14] example from real life projects is
[29:16] product recommendation combined with
[29:18] website scraping for Affiliates an AI
[29:21] chatbot can provide customer support
[29:23] match clients with the best products and
[29:26] then scrape websites like like Amazon to
[29:29] recommend the products while attaching
[29:31] affiliate links you know on Amazon you
[29:33] can get an affiliate link and earn some
[29:35] commission fee on each sale so that is
[29:37] what Affiliates start using these
[29:39] chatbots for matching users with
[29:42] products and fetching products with
[29:44] affiliate links already in real time
[29:47] basically selling them to leads right
[29:49] away and there are many smart and
[29:50] creative ways to utilize AI chatbots
[29:54] obviously the more experienced you are
[29:56] the more advanced you are in software
[29:58] development the more comprehensive
[30:01] projects you can take on and if you are
[30:03] interested in exploring more use cases
[30:05] check out my video titled top five AI
[30:07] automations to sell in 2024 where I
[30:10] cover more use cases and
[30:13] projects moving on let's review the
[30:16] entire toolkit you need to get started
[30:18] in this space my toolkit here is kind of
[30:21] default one the same tools Liam showed
[30:23] in his video 10 months ago I'll quickly
[30:26] go through them and show you how they
[30:28] changed and the new features they offer
[30:30] because now you can achieve much more
[30:33] using the same tools okay so here's our
[30:36] metrix prototyping software they are
[30:38] extremely easy to use right you can
[30:41] create a basic AI chatbot in a few
[30:43] clicks but they are still quite Limited
[30:46] in terms of customization okay chatbase
[30:49] a year ago in chatbase you could add
[30:52] documents or paste a website URL to use
[30:55] them as a knowledge base for an ahr and
[30:58] you could only deploy it to a website as
[31:00] a widget you know now in addition to
[31:03] documents and website data you can use
[31:05] notion so all your pages in notion can
[31:08] be used as a knowledge base they've also
[31:10] introduced a bunch of Integrations so
[31:12] you can deploy the chatboard to Whatsapp
[31:14] which is very useful today and you can
[31:16] also integrate it with zapier slack or
[31:18] WordPress let's now create a customer
[31:21] service AI chatbot and deploy it to a
[31:23] website using chatbase just to show you
[31:26] how quickly you can do it this is going
[31:28] to be our demo web page it's for bossar
[31:31] cosmetics and we want to deploy a
[31:33] chatbot here we can see there are no
[31:35] chatbot widgets displayed at the moment
[31:37] so let's go to chatbase and click create
[31:40] new chat button right away you are
[31:42] prompted to add your data sources you
[31:44] can add files text websites or connect
[31:48] notion this is something new right so
[31:50] let's try this out I'm going to click
[31:52] connect notion I understand here we
[31:56] select which pages to use I have my
[31:58] bossar Cosmetics knowledge base prepared
[32:00] so I'm going to allow access to that one
[32:04] by the way all of these sample resources
[32:06] such as knowledge base prompt site HTML
[32:09] and the entire presentation will be
[32:12] available for free in my school
[32:14] community and you can access it using
[32:16] the link in the video description let's
[32:18] click create chatbot it takes a few
[32:20] seconds to create all right so basically
[32:23] it is done you can check which model you
[32:25] are using um such as GPT 40 in this case
[32:28] in the activity section you have chat
[32:30] logs and analytics if you go to sources
[32:33] you can add more knowledge based files
[32:35] at any moment the connect tab here you
[32:38] can embed the chatbot to your site share
[32:41] it in a separate URL or integrate it
[32:43] into WhatsApp or any other apps that we
[32:46] discussed a moment ago in settings you
[32:49] can go to Ai and select a large language
[32:51] model you can modify the r instructions
[32:54] we have some preconfigured default roll
[32:56] and constraints here here but you
[32:58] remember that promt that we kind of
[33:00] created together according to the best
[33:02] prompt techniques let's just copy and
[33:04] paste it here and temperature unless you
[33:07] want to use it for Creative tasks like
[33:10] creative writing or ideation just keep
[33:12] it at zero now you can customize how it
[33:15] looks change the colors the icon Etc and
[33:19] you can embed or share it just make it
[33:21] public and let's quickly test it out on
[33:24] a separate page first okay hi what are
[33:27] you your products hello welcome to
[33:29] bossar Cosmetics we offer a wide range
[33:32] of high quality Beauty and skincare
[33:34] products and then it provides me with a
[33:37] list of products nice it works well now
[33:40] to actually put it on our website I'm
[33:42] going to copy this script here then go
[33:46] to the HTML of the website and paste it
[33:51] somewhere somewhere here save it then go
[33:54] back to the website refresh it and we
[33:57] have our chatboard widget displayed in
[33:59] the corner hi do you have any hair
[34:03] products yes we do have a variety of
[34:06] hair products available are you looking
[34:09] for shampoos conditioners let's say
[34:12] shampoos which ones do you have and it
[34:15] gives me options available according to
[34:17] my knowledge base in notion and that is
[34:20] how this simple prototyping works you
[34:22] you can set it up in a few minutes you
[34:24] just need to have your knowledge base
[34:26] and prompt or roll instructions prepared
[34:29] okay danta AI dant AI is also a great
[34:32] tool for prototyping a year ago you
[34:35] could use documents and websites as a
[34:37] knowledge base and you could add a
[34:39] YouTube url which would be automatically
[34:41] transcribed and used as a knowledge
[34:43] based now they have introduced Google
[34:45] Drive and Google Sheets as sources of
[34:48] knowledge for a chatbot and that's
[34:49] already a lot additionally they have
[34:52] preconfigured some popular
[34:53] functionalities the chatbot can now
[34:55] collect user data with the lead
[34:58] generation forms and book meetings using
[35:01] your calendar links this is how it looks
[35:03] like you can click create AI chart Bots
[35:06] let's call it bosar Cosmetics assistant
[35:08] click next and you can either upload
[35:12] files or URLs it could be YouTube Google
[35:14] Drive Google Sheets or website I have my
[35:17] product database in Google spreadsheet
[35:20] with product names and pricing so I'm
[35:22] going to copy this link and paste it
[35:25] here click next review and conf firm
[35:28] then create the chatbot okay now it
[35:30] should use my data for example we have
[35:33] this product name and the price for it
[35:35] is
[35:37] $4.99 let's ask what is the price for
[35:41] and paste that product name it replies
[35:44] the price is $14.99 so it is
[35:47] successfully using my Google
[35:48] spreadsheets as a knowledge base and
[35:50] that's great on top of that I love the
[35:53] user experience here the system guides
[35:55] you through the customization steps you
[35:58] can modify the appearance of the chatbot
[36:00] you can add your logo the chatbot URL
[36:03] the next step is the chatbots
[36:05] personality so there are some
[36:06] preconfigured templates for you to
[36:08] choose from or you can create your own
[36:11] prompt so I'm going to use my prompt
[36:13] again just copy and paste it here and at
[36:17] the bottom you'll see chatbot creativity
[36:20] which refers to the temperature right so
[36:22] they just named it differently but it is
[36:25] the same thing it determines how
[36:27] creative or random the responses might
[36:29] be next you can change the welcome
[36:32] message you can add some suggested
[36:34] prompts and if you choose so always they
[36:38] will be displayed here on the right
[36:39] let's say what are your products and it
[36:44] looks like this next is lead generation
[36:47] and this is really impressive listen I
[36:49] have a video where I built a lead
[36:51] generation chatbot in voice flow and it
[36:53] was quite complex there were a lot of
[36:55] steps involved in in that I also used
[36:58] mag.com to connect the chatbot with
[37:01] Google spreadsheets using web Hooks and
[37:03] I had to set up the triggers and so on
[37:05] but using dant AI you can achieve the
[37:08] Same by just checking this box and
[37:10] describing when you want the lead
[37:12] generation form to show you can allow
[37:14] the user to skip the form you can
[37:16] uncheck the option to show it at the
[37:18] start and instead describe a condition
[37:20] when the form should pop up for example
[37:24] when the user asks to be contact Ed by
[37:29] agents so whenever they ask to speak to
[37:32] a human agent the chatbot would collect
[37:35] their contact details and this way you
[37:38] generate leads right you can also add
[37:41] more Fields like phone number name email
[37:44] Etc all right and another big update is
[37:47] booking meetings you can just paste your
[37:50] calendly link and describe when you'd
[37:52] like the book meeting button to appear
[37:55] for example when the user asks for a
[37:57] meeting or you can also set it to always
[38:00] be visible um and it looks like this at
[38:03] the bottom of the chatbot window I
[38:05] really like these two options they are
[38:07] of course for paid users only so you'd
[38:10] have to upgrade to use them but as for
[38:12] prototyping software now there is much
[38:15] more flexibility they also offer some
[38:17] Integrations here you can connect your
[38:19] chatbot with WhatsApp messenger zapier
[38:23] and more and they made it very easy I
[38:26] mean they provide you with a detailed
[38:29] step-by-step integration guide so if you
[38:31] want to connect it to Whatsapp you don't
[38:34] even need to search you know for YouTube
[38:36] tutorials or something like that it is
[38:38] all here then we have chatbot Builders
[38:41] which are much more flexible you can
[38:43] Implement more advanced features using
[38:46] tools like voice flow or botpress but
[38:49] they are harder to use they have kind of
[38:51] a modular structure requiring you to
[38:54] build the chatbots workflow logic step
[38:57] by step step it's not as userfriendly
[38:59] and preconfigured as chatbase or Dante
[39:02] and to create really Advanced features
[39:05] with voice flow or botpress you often
[39:07] need to use some some web hooks or write
[39:10] a few lines of code so I'd say that
[39:13] these are low code rather than no code
[39:16] tools when it comes to building more
[39:19] advanced Solutions okay comparing the
[39:22] two B press is definitely more
[39:24] complicated and requires more technical
[39:27] back ground so I put together this
[39:29] comparison table bpress is an open-
[39:31] Source conversational AI platform which
[39:34] means it's flexible but might require
[39:37] more Hands-On work okay voice flow on
[39:40] the other hand is a no code platform so
[39:43] if you're not into coding voice flow
[39:45] might be easier to get started Target
[39:47] users bpress is geared more towards
[39:50] developers and businesses while voice
[39:53] flow targets designers product managers
[39:55] and also businesses so again it's more
[39:58] user friendly especially if you don't
[40:00] have a developer background
[40:02] customization bpress offers extensive
[40:04] customization with its modular
[40:07] architecture you get a lot of control
[40:09] here Voice Low is more limited to the
[40:12] platforms features right it's
[40:14] straightforward but less flexible
[40:17] hosting bpress is self-hosted you have
[40:19] to manage your servers Voice Low is
[40:21] hosted by voice flow so they handle the
[40:23] hosting which is one last thing to worry
[40:26] about pricing they both have free plans
[40:30] available so you can test them out right
[40:32] away overall pros and cons bpress gives
[40:35] you full control over data and
[40:37] deployment it's highly customizable if
[40:39] you have the skills but it has a steeper
[40:42] learning curve at the same time voice
[40:45] flow is userfriendly and still provides
[40:48] enough customization to be far more
[40:50] advanced than chatbase or dant however
[40:54] comparing to bpress you'd be more
[40:56] dependent on their preconfigured
[40:58] features and have less control over
[41:01] deployment so there is always this
[41:03] tradeoff you know between flexibility
[41:05] and ease of use I'm going to provide you
[41:08] with a voice flow tutorial later in this
[41:10] video so just stay tuned okay then we
[41:13] have what I call integration tools and
[41:16] there are many options however mag.com
[41:18] and zapier have proven to work well in
[41:21] our context I mean with these tools you
[41:24] can build workflow automations that in
[41:27] enhance the capabilities of your AI
[41:30] chatbots for example if you need to
[41:31] establish communication between your
[41:33] chatbot in voice flow and a third party
[41:36] tool like Google spreadsheets or a CRM
[41:39] system or many other apps you can use
[41:42] m.com to create a scenario and just drag
[41:46] and drop these apps to connect them
[41:48] instead of coding the API integration if
[41:50] we compare them I'd say they are quite
[41:53] similar in terms of what you can achieve
[41:55] just as bpress is technically Ally more
[41:57] advanced compared to voice flow in this
[42:00] case mag.com can be more complex for
[42:03] non-technical users sometimes building
[42:06] Advanced scenarios requires some
[42:08] technical background to kind of set it
[42:11] up properly other than that they are
[42:13] both drag and drop Solutions used to
[42:15] build workflow automations and both
[42:18] offer very generous free plans so if you
[42:21] want to do something like complex
[42:23] workflows requiring multiple app
[42:25] Integrations such as connecting chatbots
[42:28] to CRM systems or creating customer
[42:31] support tickets from chatbot chats I'd
[42:33] go with me.com for its flexibility and
[42:36] ability to handle more sophisticated
[42:39] scenarios but if it's simpler or more
[42:42] straightforward automations you are
[42:44] after like automatically sharing new
[42:46] blog posts on social media or
[42:48] automatically creating tasks in project
[42:51] management tools for example in jera you
[42:53] know triggered by new emails or form
[42:56] submissions for that I'd go for zapier
[43:00] for its userfriendly interface and
[43:02] really extensive library of predefined
[43:06] templates and since these tools are
[43:08] probably the most popular you can find a
[43:11] tutorial on YouTube for each of the
[43:13] automation tasks I just mentioned and
[43:16] that way you can learn them you can
[43:17] learn how to use them in no time really
[43:20] and then we have custom code option this
[43:22] is what we do as our agency and just to
[43:25] give you more insights we use nodejs for
[43:28] all our functions sure you can get the
[43:30] same results with other programming
[43:32] languages and we know python we know PHP
[43:35] and rust but we mainly stick to
[43:38] typescript and nodejs because we have
[43:41] the most experience with them okay we
[43:43] use AWS S3 to store files we deploy our
[43:47] functions to AWS Lambda of course the B
[43:49] response time is important so you need
[43:51] to reduce it as much as possible we use
[43:55] LRT as the run time in our Lambda
[43:58] functions to achieve this don't
[44:00] overthink it okay I just mentioned this
[44:02] in case you have a technical background
[44:05] and are curious what our devs use so why
[44:08] did we choose to custom code our
[44:10] Solutions instead of building them in
[44:13] voice flow and connecting to other apps
[44:15] using mag.com for example first of all
[44:18] this is basically the cheapest approach
[44:20] because you pay fewer third party margin
[44:23] Fe second it is the most flexible
[44:26] solution because are not dependent on
[44:28] what was preconfigured by the voice flow
[44:30] development team you can create any
[44:33] solution according to the client's needs
[44:35] I'll give you an example let's say we
[44:37] want to build an AI powered customer
[44:40] support chatbot that can recommend
[44:42] products based on customer needs you can
[44:45] build a chatbot in voice flow Implement
[44:47] a product recommendation algorithm then
[44:50] connect it to make.com store the product
[44:53] database in air table or Google
[44:55] spreadsheets use web hooks to connect
[44:57] your voice flow chatbot with mag.com and
[45:00] by the way I'll show you how to do
[45:02] exactly that in a moment or you can
[45:04] write custom code build an AI assistant
[45:07] and connect it via API to air table or
[45:10] Google spreadsheets or whatever it is
[45:12] you want you'll achieve the same result
[45:14] but the second option is more flexible
[45:16] for example if you wanted to add image
[45:19] recognition on top of that using B press
[45:21] or voice flow wouldn't be possible
[45:23] because they don't support image
[45:25] recognition but but using custom code
[45:28] you could just modify the code and add
[45:30] new features of course the entry bar
[45:32] here is higher because you need a
[45:34] software development skill set I must
[45:36] say though that out of all the leads
[45:39] we've had about 80% of the projects
[45:42] would not be possible to complete if we
[45:44] only used no code or low code
[45:49] Solutions now let's move on to the
[45:51] Practical tutorials we already built a
[45:54] basic customer support chatbot in chat
[45:57] base this time I'm going to make it a
[45:59] bit more advanced I'll show you how to
[46:01] build a customer support chatbot which
[46:04] will also be capable of recommending
[46:07] product listings to customers and for
[46:10] that I'm going to use voice flow as a
[46:13] chatboard builder then Google
[46:16] spreadsheets to store my product
[46:18] database and mag.com to connect voice
[46:21] flow with Google spreadsheets so that my
[46:24] chatboard could have access to my
[46:26] inventory in real time all right let's
[46:28] start with the demo here's our demo
[46:30] website I've already added our chatbot
[46:32] widget let's start a new chat how can I
[46:35] help you find the perfect product okay
[46:37] can you recommend a serum and it
[46:41] provides me with product listings these
[46:43] are the serums available in my product
[46:46] database there are buttons to visit the
[46:48] product page and to purchase each
[46:51] product also has a brief description
[46:53] next let's ask can you also recommend a
[46:59] scrub anything under $8 and yep it
[47:03] recommends the treeh hot she sugar scrub
[47:06] it's a great option within your budget
[47:08] and it's recommended only one product if
[47:11] we go to our Google Sheets where I store
[47:13] my products and check the subcategory
[47:16] column we'll find scrubs there okay
[47:19] there are two scrubs one below $8 and
[47:22] one above that's why it recommended only
[47:25] this one which satisfies my request this
[47:28] is how the whole chatbot looks in voice
[47:31] flow it's not too complicated and we are
[47:33] going to build it together from scratch
[47:36] we will use this spreadsheet as our
[47:39] product database I'll also upload this
[47:41] knowledge base which is for my fake
[47:44] online store you remember bossar
[47:46] Cosmetics once you sign up with Voice
[47:48] flow click new agent enter the agent
[47:52] name let's say bosar Cosmetics assistant
[47:56] select modality chart and select English
[48:00] and create the agent this is going to be
[48:02] our workspace okay let's delete all
[48:05] these beginner tips here and first we
[48:08] want to add a knowledge base go to
[48:11] knowledge and click add data source here
[48:15] I'll add it as a plain text select all
[48:17] of it and copy paste it here now it has
[48:20] some information about bossar Cosmetics
[48:23] let's go back to workflows click edit
[48:26] work close and here we will start
[48:29] building I made it extremely easy for
[48:31] you every step is described in a word
[48:34] file that I will also attach in my
[48:35] resource Hub in school Community it
[48:38] details every step and when I say every
[48:41] step I mean if if it says talk text it
[48:45] means you go to talk and then text so
[48:49] there should be no confusion at all also
[48:51] you have all the text and code that you
[48:53] can just copy and paste such as this
[48:55] welcome message I'm going to generate a
[48:58] few more variants and this is how
[49:00] detailed this guide is so feel free to
[49:03] use it okay so go to listen buttons then
[49:06] click no match and create a path
[49:09] connected to a new blog it should be
[49:12] logic set name it set AI question and
[49:16] apply to Let's create a new variable
[49:19] name it question and as a value select
[49:22] last utterance which is the reply from
[49:26] from the the user in the chat okay
[49:29] connected to the next block AI set AI
[49:33] select AI model as a data source and
[49:36] paste this prompt classify whether this
[49:39] user is asking for a product
[49:41] recommendation or not if they are asking
[49:43] for a product recommendation say yes if
[49:46] not say no apply this to a variable
[49:49] recom and create the variable next add a
[49:53] logic block choose condition set it so
[49:56] that if the variable recom contains yes
[50:00] it will go to the product recommendation
[50:02] algorithm if no other words no match
[50:05] create a path and it will go to the AI
[50:08] text block so the next block is AI set
[50:12] AI here keep AI model as a data source
[50:15] again and paste this prompt here is what
[50:18] the customer has requested we have our
[50:20] question reply to this question
[50:22] according to the knowledge base if you
[50:24] don't have an answer refer to support at
[50:27] postar cosmetics.com and then create a
[50:29] new variable let's call it recom text
[50:32] and apply it let's label it AI text and
[50:35] now go to talk text drag and drop it
[50:38] here and select our variable Recon text
[50:43] then connect it back to the first block
[50:45] okay so this part is done it can already
[50:49] provide customer support and determine
[50:51] if the user is asking for product
[50:53] recommendation or not let's mark it with
[50:55] one color okay the idea here is to check
[50:59] if a product question is asked if yes it
[51:04] will make request to me.com if not it
[51:07] will return us back to the first blog
[51:10] okay let's run a test welcome to bossar
[51:14] Cosmetics when are you open the response
[51:18] is thank you for reaching out to bossar
[51:20] Cosmetics our store hours are Monday to
[51:23] Friday blah blah blah so according to
[51:25] our knowledge B it replies correctly now
[51:28] we need to build the product
[51:30] recommendation part let's add a new
[51:32] block logic set here we need to set our
[51:36] Google Sheets variables so this part of
[51:39] the assistant will be responsible for
[51:41] running air table request sorry I meant
[51:45] m.com request not air table here we need
[51:48] to set Google spreadsheets IDs let's add
[51:51] a few sets the first one is applied to
[51:54] spreadsheet ID let's create this
[51:58] variable okay then go to your Google
[52:01] spreadsheets URL and the spreadsheets ID
[52:04] is this part after d slash up until
[52:08] sledit paste this ID as a value here
[52:11] with a quotation marks the second one is
[52:14] the uh sheet ID create the variable and
[52:17] you can find this ID in your Google
[52:19] spreadsheets URL again after GID equals
[52:23] so in our case it's zero next go to log
[52:26] iic set this will be our main Google
[52:29] spreadsheets logic okay it will set the
[52:32] number of Google Sheets row responses
[52:35] apply to number of responses okay create
[52:39] variable and I want to set it to four to
[52:42] have only up to four product listings in
[52:45] the chatbots reply okay then go to AI
[52:48] set AI drag and drop it
[52:51] here this will be our Google
[52:53] spreadsheets query choose AI model as a
[52:56] data source and paste my prompt which is
[52:59] convert the following query to Google
[53:01] charts query language if there is no
[53:03] valid query reply with there is no valid
[53:06] query the query should only at Max
[53:09] include the product category create a
[53:11] variable spreadsheets query and then go
[53:14] to the prompt settings and paste my
[53:16] system prompt from the guide obviously
[53:19] you'd have to modify it according to
[53:21] your needs according to your product
[53:22] database and your context but overall
[53:25] these instruction describe how to
[53:28] convert user queries into queries for
[53:30] Google Sheets we provide the column
[53:32] names according to the columns in Google
[53:35] spreadsheets it should be a b c d we
[53:38] list the products product categories and
[53:42] subcategories and provide a few more
[53:44] instructions here for example if they
[53:46] ask for something that is not listed
[53:49] just assign a product that is close to
[53:51] what they want for example they ask for
[53:53] a shower gel just assign body wash
[53:56] subcategory since it is close to
[53:58] accomplishing the body washing purpose
[54:00] of the product you are only to answer
[54:02] with the query for example input do you
[54:05] sell any serums assistant like the
[54:07] output should be only the subcategory
[54:10] all right the next text block is just a
[54:12] logic block so go to logic condition if
[54:15] query that would be a new variable you
[54:18] need to create it if it contains no
[54:21] valid query go to the AI text block the
[54:24] one we created here so it will go back
[54:27] to the loop okay and if no match if the
[54:30] query is valid then we want to make a
[54:33] request to make thatc so let's add
[54:36] another block which would be DAV API and
[54:40] in this block we'll configure the API
[54:43] call to mag.com we need to set what we
[54:46] are going to pass to mag.com so first we
[54:49] need to switch this to post now I want
[54:52] to add the body we are going to send the
[54:54] query which will be the spreadsheet
[54:57] query it was set in the second block
[54:59] here then we want to add the spreadsheet
[55:02] ID the one we set in the first block and
[55:04] also the sheet ID which was also set in
[55:07] the first block then we have capture
[55:10] response let's set the response and
[55:12] apply it to formatted response variable
[55:16] and we need to create this new variable
[55:18] okay if it fails we need to add a new
[55:20] text block and say something like sorry
[55:23] something went wrong please try again
[55:27] and I'll generate more variations here
[55:29] let's also Mark this block as failed and
[55:33] give it a red color if it succeeds we
[55:36] will continue okay but for now let's set
[55:39] up a web hook for mag.com go to mag.com
[55:42] sign up and create a new scenario here
[55:46] the first component should be custom web
[55:49] hook create a new web hook let's name It
[55:52] bossar Cosmetics voice flow and save
[55:56] okay it will be waiting for variables
[55:59] select this URL copy it and paste it
[56:03] into Post in our API blog in voice flow
[56:08] now let's send our variables to mag.com
[56:10] click run run
[56:13] test say um recommend me a serum it
[56:19] should go through the whole logic here
[56:22] it should be successful but we haven't
[56:24] built the following block yet however in
[56:27] mag.com our data structure is
[56:30] successfully determined the next block
[56:33] should be Google Sheets so scroll down
[56:36] and select search rows Advanced okay
[56:41] connect your Google account here and
[56:43] leave enter manually we need to select
[56:46] our variables so spreadsheet ID sheet ID
[56:51] and query and set the maximum number of
[56:56] return rows to 10 okay by the way for
[57:00] mag.com I will also attach this guide so
[57:03] you can follow it step by step without
[57:05] any trouble now the data we receive from
[57:08] Google spreadsheets will be aggregated
[57:11] into Json and then the Json string will
[57:15] be returned to our voice lowboard so add
[57:19] a
[57:20] Json aggregate to Json The Source model
[57:24] should be your Google Sheets here here
[57:26] data structure I just select product cuz
[57:29] I have this preconfigured data structure
[57:33] in your case you'll have to configure it
[57:35] so you'll have to click add and add
[57:38] items according to your column names
[57:41] such as product name then add item again
[57:45] category you want to add all these items
[57:47] one by one now you see I don't have any
[57:50] values populated from my spreadsheet yet
[57:53] so let's run the whole thing again to
[57:55] populate some values click run once run
[57:59] anyway run the whole thing again okay
[58:03] recommend me a serum so it's going to
[58:07] query the Google Sheets now and if I
[58:09] switch back to mecom and go to that Json
[58:13] component I should have these values for
[58:15] my columns populated product name
[58:18] category subcategory description price
[58:21] and image link the last block is web
[58:24] hook response it's going to send the
[58:25] Json string back to our voice flow bot
[58:29] okay for body select Json string and I
[58:32] also want to add some custom headers
[58:35] here so content
[58:38] type and application Json click okay and
[58:43] make.com scenario is now set up let's
[58:46] reset test run it and say recommend me a
[58:52] serum it's going through the flow and in
[58:55] me.com everything is initialized and
[58:58] finalized successfully just make sure
[59:01] that if you go to scenarios this
[59:03] scenario is turned on okay to make it
[59:06] work the next block is dev then
[59:09] JavaScript this block takes the Google
[59:12] spreadsheet data and converts it into
[59:16] variables basically this part of the
[59:18] system is to run our make.com request so
[59:22] let's mark it with one color for the
[59:24] JavaScript block you need to enter the
[59:26] JavaScript code here just copy and paste
[59:29] it from my guide so we get the response
[59:32] from meg.com then product count
[59:35] determines how many products it returned
[59:38] if more than zero then we set the
[59:41] variables here very repetitive code to
[59:44] be honest but you know for this
[59:46] structure you'd have to do it obviously
[59:49] modifying it using your names of of your
[59:52] own columns if it fails go back to the
[59:56] AI text block and start over if it
[59:59] succeeds we add a new block which is
[60:02] logic condition and this part is just to
[60:05] set the logic and make it display the
[60:07] right amount of product listings
[60:09] according to the amount of products we
[60:10] got in the response from mag.com the
[60:13] first one is zero let's create a
[60:15] variable product count if the product
[60:18] count is zero create One path then
[60:22] another condition if product count is
[60:25] one and the same we do for two three
[60:29] four and then no match create a path so
[60:34] if mag.com returns two products then it
[60:37] will go to two product listings if four
[60:41] then it will go to four and if zero then
[60:44] we'll send it back to our AI text
[60:46] response and kind of close the loop else
[60:49] means that it is not 0 1 2 3 4 so it is
[60:54] five or more and in that case we want to
[60:57] display also four products because
[60:59] that's the maximum amount of product
[61:00] listings we want to display so I'll
[61:03] connect it to the same product listing
[61:04] blog as if it was four products next we
[61:08] need to create four blocks it will be AI
[61:12] set AI the goal of these blocks is to
[61:14] create the follow-up messages to support
[61:17] the product listings right to to
[61:19] describe the suggested products so text
[61:22] one select AI model as a data source
[61:25] then just paste my prompt here here's
[61:28] what the customer has requested our
[61:31] variable for question here's the query
[61:33] that will be ran spreadsheet query
[61:36] here's the product recommended product
[61:38] one name which is our variable for for
[61:40] the first product if the product
[61:42] recommended is not what they asked for
[61:45] please tell them that we don't have what
[61:47] they are looking for but we found this
[61:50] as a close alternative and we want to
[61:52] apply it to our variable recom text to
[61:56] be more specific you can provide a
[61:57] system prompt here and I like to do that
[62:00] usually something like your job is to
[62:02] help the customer understand the
[62:03] products they were recommended your
[62:06] answers need to be short and concise Max
[62:08] one to two sentences above this message
[62:11] will be the product listed so we don't
[62:13] need to ask if they want to see them and
[62:15] don't ask any queries just to be safe
[62:18] all right duplicate it three times and
[62:21] add more recommended products here is
[62:24] the first product recommended the second
[62:26] the third and the fourth and then just
[62:29] modify the second and third blocks here
[62:31] accordingly the last step for the whole
[62:33] system is to display the product
[62:35] listings add a new block talk then
[62:38] Carell here you want to switch to link
[62:41] and create a few variables the first one
[62:44] is product image created the second one
[62:48] should be product name okay and the
[62:51] third one product price then we can add
[62:55] some buttons here for example visit
[62:58] product and if you have a website with a
[63:01] product listing you should go to actions
[63:04] select open URL and paste your url the
[63:09] second button can be purchase and again
[63:11] you can add your url here if you have a
[63:14] website then go to talk text drag and
[63:18] drop it here select our variable ROM
[63:24] text that's the one the AI model
[63:26] generates in the previous block here
[63:29] according to our instructions right it
[63:32] will complement the product listing with
[63:34] a
[63:35] description okay then go to listen
[63:39] button drop it here name it like let's
[63:43] start over this is just to complete the
[63:46] loop so actions go to block search for
[63:51] start and it will bring the user back to
[63:54] the first block just duplicate this
[63:57] adding more product cards according to
[63:59] the product number every time you'd have
[64:01] to create new variables like product two
[64:04] image product two name and product two
[64:06] price Etc and once it is all done our
[64:10] chatboard is basically ready to be used
[64:12] okay this is how our whole system looks
[64:16] like let's click run recommend the best
[64:20] serum you have and let's see how it
[64:23] works it is going through this steps and
[64:26] this is how the output looks like we
[64:28] have the product listing two buttons
[64:31] then a brief description of this serum
[64:35] and a button to start over if I click
[64:37] this button it will begin the flow from
[64:39] the start so we don't have to build it
[64:41] from scratch you can just modify it
[64:43] according to your needs I will attach a
[64:44] template to this chatbot in my resource
[64:47] Hub so you don't have to actually build
[64:48] it from scratch you can just import it
[64:51] and modify according to your needs many
[64:53] people ask me how to use the templat so
[64:56] let me just quickly show you in voice
[64:58] flow click on the icon in the top right
[65:01] corner to import the template upload the
[65:05] template and you'll be able to edit the
[65:08] workflow okay for make.com create a new
[65:11] scenario click on the three dots and
[65:14] select import blueprint upload my
[65:18] template in Json format and you'll get
[65:20] access to my scenario this system is
[65:22] quite basic it can only search by
[65:24] categories and subcat atories and sort
[65:27] by price but later in this video I'll
[65:30] show you a b that can actually analyze
[65:33] product descriptions and evaluate
[65:35] customer needs and then match the
[65:37] relevant products to customer needs now
[65:40] pay attention this is important good
[65:42] news if you really Master chatboard
[65:45] Builders like voice flow and integration
[65:47] tools like mag.com you can already do a
[65:50] lot you can provide real value and there
[65:53] are many tutorials on YouTube on how to
[65:55] to use these tools in including my
[65:58] channels so you'll have enough resources
[66:00] to learn from but here's the bad news
[66:03] since there are so many tutorials and
[66:05] these tools are userfriendly requiring
[66:08] no code no background in development a
[66:11] lot of small and medium-sized business
[66:13] owners would rather watch the same
[66:15] tutorials and do it themselves instead
[66:18] of paying you a few th000 for
[66:21] implementation every second lead that
[66:23] books a call with us always says well I
[66:26] am technical enough I can use voice flow
[66:28] and mag.com but when it comes to code
[66:32] that's where I'm stuck so my point is
[66:36] that if you weren't limited to no code
[66:38] tools and low code tools and you could
[66:41] build some kind of custom solution to
[66:44] fit the customers's need then you would
[66:47] have a great competitive Advantage now
[66:49] the big question is where can you learn
[66:52] to build custom code for these Ai and
[66:54] automation Solutions
[66:56] well you could spend a year learning
[66:58] Python and then even more time figuring
[67:00] out how to apply those skills to these
[67:03] Solutions but a more efficient way would
[67:06] be to take some kind of a coding crash
[67:08] course specifically designed to give you
[67:11] the Knowledge and Skills to build
[67:14] exactly these kind of AI and automation
[67:16] solutions that you can sell as an AI
[67:19] automation agency and this is exactly
[67:22] what we are going to offer we are going
[67:24] to launch the AI Fellowship a community
[67:27] program consisting of three pillars the
[67:30] first one is AI automation coding crash
[67:33] course we've noticed that if you learn
[67:35] how to build and adopt 10 to 15
[67:37] solutions for different businesses you
[67:39] can handle about 80% of projects they
[67:43] really repeat a lot and since we are
[67:45] doing it we know which Solutions are in
[67:48] high demand that's why we are putting
[67:50] together a curated crash course focusing
[67:52] on AI and automation solutions that are
[67:55] currently being sold it is the 80/20
[67:58] rule you need just 20% of the effort to
[68:02] achieve 80% of the results and our team
[68:04] of developers is preparing the modules
[68:06] right now to provide you with that
[68:08] crucial 20% of technical knowledge most
[68:12] relevant to our field then you also need
[68:14] to know how to sell these Solutions you
[68:16] can save many months of your life by
[68:19] learning from our trials and errors so
[68:21] along with the coding course we'll
[68:24] provide AI automation agency coaching
[68:27] this is a complete guide on how to start
[68:29] and scale your business you'll learn how
[68:31] to generate leads and close them you'll
[68:33] get an entire toolkit for running an
[68:36] agency including email templates
[68:38] contract templates how to price Your
[68:40] solution and basically everything from A
[68:42] to Z and of course the most valuable
[68:44] part of the coaching pillar is the
[68:47] lessons and tips that come only from
[68:50] real life experience and what I believe
[68:52] to be the most important part of the
[68:54] whole AI
[68:55] program is the community you can find
[68:59] all the information and knowledge about
[69:01] coding and sales on the internet we just
[69:03] save you a ton of time and money by
[69:04] providing curated modules but what is
[69:07] not so easy to access is a community of
[69:10] like-minded people working towards
[69:13] similar goals and that's the third
[69:15] pillar of our program we'll have
[69:17] masterminds a closed Discord Community
[69:20] will introduce a matching system so if
[69:22] you are a salesperson looking for a
[69:24] developer or vice versa will match you
[69:27] and the support you'll get from other
[69:29] participants is invaluable instead of
[69:32] working alone you'll be a part of a
[69:34] group where someone is just a step or
[69:36] two ahead of you and has faced the same
[69:38] challenges the community is truly the
[69:41] the most important part of the program
[69:43] I'll attach a link to the AI Fellowship
[69:45] where you can sign up for the waiting
[69:46] list the first 50 people on the list
[69:49] will get a 50% discount on the entire
[69:52] program so sign up now and I'll provide
[69:55] more more details soon I'm going to show
[69:57] you two projects with custom code AS
[69:59] examples of what you'll be capable of
[70:03] once you complete the course but before
[70:05] that let's quickly review the top large
[70:07] language models available
[70:10] today it is important to know at least
[70:12] the top ones because they each have
[70:15] their pros and cons for different tasks
[70:17] and you want to use the right model for
[70:19] the right task overall there are three
[70:21] models we usually test to pick the best
[70:24] one for a project project Google's
[70:26] Gemini anthropics clae and open AI GPT
[70:30] I've put together this table to compare
[70:32] the latest versions Gemini 1.5 Pro clae
[70:36] Sonet 3.5 and GPT 40 I'm going to refer
[70:40] to them as clae Gemini and GPT 40 not to
[70:43] repeat every time Sonet 3.5 Gemini 1.5
[70:46] Pro and so on so when it comes to the
[70:49] context window clae offers 200,000
[70:52] tokens Gemini 1 million making it
[70:55] perfect for handling extensive data sets
[70:57] and kind of long documents and GPT 40
[71:00] provides
[71:02] 128,000 tokens which is more than enough
[71:05] for many tasks but the smallest of the
[71:08] three models here talking about speed
[71:10] according to our test CLA is faster than
[71:12] GPT 40 but slower than Gemini so GPT 40
[71:17] is currently the slowest among the three
[71:19] looking at costs Claud charges $3 per
[71:23] million input tokens and $15 per million
[71:26] output tokens Gemini 350 for the input
[71:29] tokens and 1050 for the output tokens
[71:32] GPT 40 is the most expensive with input
[71:34] tokens costing $5 per million and output
[71:37] tokens at $15 per million GPT 3.5 turbo
[71:41] is 10 times cheaper by the way so we
[71:44] usually test it if we can achieve the
[71:46] same results um the same output
[71:49] performance with GPD 3.5 turbo also
[71:52] Gemini is the cheapest only if the
[71:54] contact window is up to
[71:56] 128k tokens once you need to use more it
[71:59] becomes the most expensive one with $7
[72:03] for input and $21 for output tokens
[72:06] overall when choosing between these
[72:08] models try to consider their unique
[72:10] strength and match them to your specific
[72:12] needs if you deal with large data try
[72:16] Gemini 1.5 Pro for detailed and precise
[72:20] tasks especially in legal and you know
[72:22] kind of educational Fields try using clo
[72:25] Sonet 3.5 and for a flexible model that
[72:28] performs well across different tasks GPD
[72:31] 40 is a good choice even though it costs
[72:34] more but take this with a grain of salt
[72:37] okay even though GPT 40 might seem like
[72:40] the worst option right based on the
[72:42] features and pricing alone it's not that
[72:44] straightforward in fact we use GPT 40 or
[72:47] GPT 3.5 turbo for 80% of our projects
[72:51] the only way to find the best model for
[72:53] your task is through testing for example
[72:56] for one of our projects we needed large
[72:59] context window theoretically Gemini's 1
[73:02] million token limit should have worked
[73:05] but it returned error in practice and we
[73:08] just couldn't use it this is just
[73:10] something from our experience for you to
[73:12] keep in mind make sure to test it
[73:14] properly now I'll show you a simple code
[73:17] to build AI chart Bots using these
[73:19] different models and to make it more
[73:21] interesting these Shard Bots will be
[73:23] able to also Rec recognize attached
[73:26] images on top of customer service this
[73:28] is our Gemini bot it is in ret I am
[73:31] going to share this code as a template
[73:33] in my resource Hub in school so you can
[73:35] copy it and play around or feel free to
[73:37] steal it I don't care in this project we
[73:40] have only three files Gemini service JS
[73:44] index JS and utility service JS this is
[73:48] just a code overview okay so don't worry
[73:51] if you don't fully understand it the
[73:53] goal is just to show you the logic of
[73:55] how it works when you join our AI
[73:57] fellowship program we guarantee that by
[74:01] the end of the course you'll be able to
[74:03] code chatbots like this one on your own
[74:06] so the first file is index JS it
[74:08] contains our endpoint where requests are
[74:11] made SL chat is our rout for requests
[74:15] from basically any chat in use be it a
[74:18] WhatsApp chat or web chat doesn't matter
[74:21] it expects two Fields chat ID and
[74:24] message then we have the validation if
[74:28] all the data is passed through and once
[74:31] validated it is passed to thees function
[74:34] which is located in Gemini service JS
[74:37] then in Gemini service JS The Bard
[74:41] creates a local database stores the
[74:44] message history or loads it if it
[74:46] already exists and it creates a new
[74:49] message and sends it to the chat then it
[74:52] receives the result stores it in the
[74:54] database and returns it the utility
[74:57] service file has only one function that
[74:59] uploads and formats an image into the
[75:01] base 64 format which is expected by
[75:04] Gemini's Library that's pretty much it
[75:07] if you need to understand this better
[75:08] here's a tip pause the video make a
[75:11] screenshot of the code upload it to CH
[75:13] GPT and ask for any clarifications you
[75:16] need obviously to make it work you need
[75:18] to add your API key from Google AI
[75:21] Studio to actually connect it to Gemini
[75:24] so go to secrets and set your API key
[75:28] for that just go to Google AI Studio the
[75:31] URL is AI studio.
[75:33] google.com/ apppp API key you need to
[75:37] log in and here you can create your API
[75:40] key create API key in new project and
[75:44] here it is just copy it and then paste
[75:47] it in ret as value once done click run
[75:52] I'm going to copy the dev URL from here
[75:54] and and go to My Demo page we just
[75:58] quickly created this page for the
[76:00] purposes of this video just you know to
[76:02] demonstrate how the chatbot works so I
[76:05] go to settings enter the URL from repet
[76:08] I need to add slash chart CU that's our
[76:12] end point okay save it and let's try it
[76:15] out hi it replies back now let's upload
[76:19] an image for example this one and ask it
[76:23] to list the objects in this photo okay
[76:29] give it a second and it provides me with
[76:31] the objects it can see in the photo we
[76:34] can also ask additional questions such
[76:36] as I like the keyboard tell me about it
[76:39] and it responds as a general model would
[76:41] usually respond that it's hard to say
[76:44] anything specific about the keyboard
[76:46] without more information however based
[76:49] on the image I can make some general
[76:51] observations and it provides me with
[76:53] specifics such as type lay out color and
[76:56] material all right now let's look at the
[76:58] cloud board don't forget to set up
[77:01] Secrets since we are using clo this time
[77:04] you need to go to entropic so console.
[77:07] antrop
[77:09] docomo you can sign up with your Gmail
[77:12] account and you'll get some free credits
[77:14] to get started just click on get API
[77:18] Keys then click create key give it a
[77:21] name say test key and click create key
[77:26] copy your key from here and paste it in
[77:30] ret as value in secret okay this Cloud
[77:34] board has a similar structure index GS
[77:37] is identical to what we had with Gemini
[77:40] it calls thees function which is this
[77:42] time located in CLA service JS instead
[77:45] of Gemini service JS utility service JS
[77:48] is also the same it uploads an image
[77:51] then in cloud service JS we create a
[77:53] database create create a new message
[77:56] then send it to an Tropic receive the
[77:59] response store it in the database and
[78:02] return it to the user okay let's test it
[78:06] out I'll click
[78:07] run copy the dev URL from
[78:11] here switch to My Demo page go to
[78:15] settings and paste the URL again don't
[78:19] forget to add slash chat Okay click save
[78:23] and now I should be able to try out hi
[78:25] we get the response from the bot now I
[78:28] will attach the same image as before and
[78:30] ask it
[78:32] to list the objects in this photo give
[78:36] it a few seconds to process and and it
[78:39] provides us with the output the bot
[78:42] recognizes the objects well it even
[78:45] recognized that the notebook in the
[78:47] image had notes written on the cover
[78:51] which is not readable for a human
[78:52] without zooming right all right let's
[78:55] ask it um tell me more about this
[79:00] keyboard and once processed it replies
[79:02] well it even identified that the
[79:04] keyboard is most likely an apple magic
[79:07] keyboard and provided many details so it
[79:10] works great I think it's even better
[79:12] than Gemini in this case that's why you
[79:14] should always test and compare the
[79:16] outputs those were clae and Gemini for
[79:19] AI assistant using gp40 I have two
[79:22] separate videos one for a General
[79:25] assistant and another with image
[79:27] recognition capabilities so be sure to
[79:29] check them out as well and for the final
[79:31] chatbot today I'll show you a bit more
[79:33] advanced solution this chatboard again
[79:36] it is going to be our online Beauty
[79:38] Store consultant for bossar Cosmetics
[79:41] this time instead of building it in
[79:43] voiceflow and make.com I'll use custom
[79:46] code and in addition to product
[79:48] recommendations it will also be able to
[79:51] recognize images so I'll basically
[79:53] combine everything we did today Customer
[79:56] Service Plus product recommendations
[79:59] plus Vision capabilities let's start
[80:01] with the demo right away to give you an
[80:02] idea of how it works and then I'll break
[80:05] down the code this ret template will
[80:08] also be available in the resource Hub so
[80:10] you can use it do not forget to set up
[80:12] your secrets this time we use GPT 4 all
[80:17] so you need to head to platform.
[80:20] open.com go to API keys and create your
[80:24] secret key once done click run wait for
[80:27] the dev URL copy it and paste it on the
[80:31] test page without this slash in the end
[80:36] and let's begin the conversation with
[80:38] the
[80:39] bot hi I need skin care it takes a few
[80:44] seconds to process the request and it
[80:47] asks me to provide a bit more
[80:49] information about your skin type and any
[80:52] specific concerns you have this will
[80:54] help help me to make more tailored
[80:56] recommendations because remember in our
[80:58] prompt we instructed the bot to find out
[81:01] about customer needs first and then
[81:04] recommend tailored products and this is
[81:06] what it does okay let's say oily skin
[81:09] and upload a photo of a face send it and
[81:14] wait for the response so it says based
[81:17] on your needs and our available products
[81:19] here are some recommendations for
[81:22] managing oily skin and address ing acne
[81:25] first of all it correctly understood
[81:27] that the customer needs products for
[81:29] managing oily skin and addressing acne I
[81:32] mentioned the oily skin but I never
[81:35] mentioned acne it recognized that purely
[81:38] from the image okay secondly it searched
[81:41] our database for the available products
[81:45] and then provided the user with the
[81:47] relevant options in the end it also
[81:49] provided a summary and a short
[81:51] description for each of the suggested
[81:53] products so let's ask it something else
[81:56] I'll say excellent also recommend a
[82:00] blush for me it is now searching our
[82:02] database for blushes and provides us
[82:06] with the relevant products along with
[82:08] the descriptions also I need a scrub and
[82:13] it suggested two scrubs there is a
[82:15] button to view product but since we
[82:17] don't have a website it doesn't redirect
[82:20] anywhere but obviously if we had a
[82:22] website it would take us to the product
[82:24] purchase page that's it for the demo now
[82:27] let's go over our code we tried to keep
[82:30] it very simple without extra functions
[82:32] for this project the product database is
[82:35] stored in an Excel file to avoid
[82:37] complicating the code with you know
[82:40] Online requests and additional functions
[82:42] in the assistant so this is just to
[82:44] simplify the process if you want to test
[82:46] it with your own product database you
[82:48] need to delete this product XLS and
[82:51] upload your own file naming it the same
[82:54] way way and when we receive a request we
[82:57] search this Excel file for the necessary
[83:00] information okay index JS just like in
[83:03] previous projects it's almost the same
[83:06] the difference is that when we receive a
[83:08] file we create an open AI file for that
[83:12] we have the upload image function and it
[83:14] is located in open AI service JS so it
[83:18] downloads the image using a URL sends a
[83:21] request to open AI purpose Vision this
[83:24] is very important to make it work in the
[83:26] assistant and then deletes the
[83:28] downloaded image and Returns the file ID
[83:31] which we received from open we then use
[83:34] this file in the message that is added
[83:36] to a thread okay here we have the same
[83:39] thing as in previous projects it starts
[83:42] with creating an assistant using the
[83:44] create assistant function this function
[83:46] is also located in open AI service JS
[83:50] also the instructions here are quite
[83:52] extensive you remember our prompt was
[83:54] quite big right so they are stored in a
[83:57] separate file instructions txt here it
[84:00] reads this file in the second line of
[84:02] code and also here we have the names of
[84:06] our product database and knowledge base
[84:08] really just check out my video on how to
[84:10] integrate gp40 assistant to a website I
[84:13] covered this whole structure there I
[84:16] covered what assistant Json is for so
[84:19] you'll have more understanding if you
[84:20] watch that video If we don't have the
[84:23] assistant Json file in the project it
[84:25] will create and save one after it gets a
[84:28] positive response from open AI initially
[84:30] we create an open AI file for the
[84:33] database and store it in a vector
[84:35] database I hope you remember what Vector
[84:38] database is I discussed it in the
[84:40] chapter about understanding AI chatbots
[84:43] if you need you can go and rewatch it
[84:46] then we create a products file and we
[84:48] will use its ID in the code interpreter
[84:51] here there are different tools available
[84:53] at openai for for example file search or
[84:56] codee interpreter so for the knowledge
[84:58] base it uses file search you remember
[85:00] how it works right the vector database
[85:03] the chunking all of that so it retrieves
[85:05] the relevant chunks of text from the
[85:07] knowledge Base According to the user's
[85:09] input and then for the product database
[85:13] it uses code interpreter it will search
[85:15] for the relevant products in our Excel
[85:18] file okay and here once the assistant is
[85:21] created it is saved in assistant Json
[85:24] file instructions txt file contains our
[85:27] prompt okay the one we created in this
[85:29] video and we just added some Specific
[85:31] Instructions here at the end to ensure
[85:34] the assistant Returns the products data
[85:36] that it found in our Excel file using
[85:39] Json format this makes it easier to
[85:41] display the product listings and that's
[85:44] it this is a simplified process just for
[85:47] the purposes of this video if it were a
[85:49] real project we'd make it more complex
[85:52] and definitely more reliable but I just
[85:55] wanted to give you an idea of how more
[85:58] advanced and custom coded AI chart Bots
[86:01] look guys if you manage to understand
[86:03] how this code Works you're probably in
[86:05] top 1% of viewers it would take numerous
[86:08] videos to actually teach you how to
[86:10] write code like this this isn't
[86:13] something you can learn from a you know
[86:15] a quick 15minute tutorial that's why we
[86:17] invite you to our AI fellowship program
[86:20] you'll get a complete course on this and
[86:22] by the end of it you'll be very
[86:24] comfortable building projects like this
[86:26] one other than that if you watched and
[86:29] understood this video till the end you
[86:32] can be proud of yourself you are now
[86:34] ahead of the majority of people who are
[86:36] interested in AI now you have a full
[86:39] understanding of what it takes to build
[86:42] these Solutions which isn't as easy as
[86:44] it might initially seem right my goal is
[86:47] not to sell you this idea but if you are
[86:50] serious and ready to commit you can make
[86:52] a lot of money if you start start now
[86:54] you can still be early enough to
[86:57] leverage this opportunity and once you
[86:59] learned how to build Ai chatbots and
[87:01] other AI Solutions and workflow
[87:03] automations you need to learn how to
[87:05] sell them the best way if not the only
[87:07] way to do this is through practice to
[87:10] get more practice you need more sales
[87:13] goals you need more leads right cold
[87:15] goals don't work here not for me not for
[87:18] other AI agency owners we actually
[87:20] discussed this recently and everyone
[87:23] agrees that it does doesn't work just
[87:25] yet you need to generate warm leads I
[87:28] have a video on how to start an AI
[87:30] automation agency where I break down
[87:32] step by step how to start and generate
[87:35] the first leads the next video on the
[87:37] channel will be the second part of that
[87:39] video with more insights and specific
[87:41] metrics I've gathered over a few months
[87:44] so make sure to subscribe and not miss
[87:47] it long story short at this stage the
[87:50] best way to get warm leads is through
[87:52] generating content putting out value
[87:55] helping people and showing your
[87:57] expertise at the same time that is what
[87:59] I'm doing today and I hope you'll
[88:01] consider it as well thank you very much
[88:03] for watching and I'll see you soon bye