welcome everyone I'm very very happy very excited for today's uh webinar it's one of a series of webinars we are conducting with our partners speak about automation using automation within your business to empower your business using AI within your automation as well today I have uh TB from make it future who has amazing amazing experience with AI and automation before I go into an introduction we're going to go through a very very quick introduction but before that dy I know you achieved some amazing results with the use case you're going to be sharing with us today can you just in a few seconds share us share with us the amazing results that you achieved yeah for sure Le thanks a lot for inviting me on this webinar hello hello everybody um I'm I'm TB um this is I can say we saved about one man of work or one man of one person of work per month right so we can calculate that about 150 hours of work per per month and if we look at from a saving perspective think about about 30k saving depending on the region or depending on your uh country right that might be different but you the potential is there right and and we're talking about to be honest about the small business here and I I'll describe that more in the next slid awesome that's that's that's amazing I think this is definitely the value that we are looking for so one one part is the saving hours but the other part is the business impact right and the business impact can be translated into Revenue but also can be translated into customer growth and customer experience I think this something we're going to be discussing uh today all right so I'm going to walk you through the agenda so the first thing we're going to go through very very quick uh introdu C then we're going to go through the the introduction to building a chat but using Ai and then we're going to go into make we're going to show you TB going to show you how exactly he built the uh the chat uh the chat bot this is a very practical webinar we're going to go into the details how we built uh how we build that all right so very quick introduction about make I hope a lot of you who are here you are already familiar uh with make make is an AI uh sorry is an automation platform uh that also enables you to uh use AI enables you to connect apps uh uh together there are thousands and thousands of people who have been using me uh for years if you're not uh using me yet uh I hope this webinar today will show you uh the value of uh of using make TB can you give us a very quick Qui introduction about make it future sure um make it future and I I think you already kind of cover this if you want maybe to in a nutshell do you want to cover this first level uh I think I think we already we already covered uh yeah let's let's move on so yeah my my testimonial to make okay is the platform I use the most as a no code platform so guys whoever is on this call take 15 minutes go test make I can tell you it's going to change the way you think about automation and business so who's Makey future we we've started with with make back in 2018 actually before we started Makey future we we spend a lot of time automating and right now we are an automation agency with 26 people and we're providing Automation Services on no code and and other PL like no code platforms like make and others like air table and hopspot and we're doing this from our development center in Romania but we also have a presence in Germany France and we're actually working on all over the globe we have customers in Israel we have customers in North America we have in Canada we have in Australia and so on so all over the all over the globe um yeah you you can see here some reference from us not so important go to our website and you can find all this information awesome brilliant thank you very much uh thank you very much uh TB now let's go into the exciting part so make it future is an automation agency right but how did it start at the beginning because I know you had a quite a very different start in business yeah and that's that's that's kind of the the nice part of nowadays of working with no code tools you don't really have to be like a coder you can start doing digitalization and you can do um Pro automation with tools like make back in 2018 we I was running uh we started a company together with my co founder pet we started a company called Blue coffee roast R our main um goal was to ship roast coffee and ship it to our customers around Romania and Europe and we had a e-commerce based on prea shop and one of our big challenges was how do we do retargeting how do we understand how uh which customer do we need to call for support which customer have problems how do we aggregate all the data and we discover hops spot at that time and we wanted to apply the CRM and the marketing um plans on on our e-commerce but the our Presta shop had no connection to HubSpot and by searching and searching to find the solution to integrate we discovered integromat and to keep the story short six months later all our processes in the business were automated okay and we realized in about six months that there's a much bigger opportunity in doing business Automation and we decided to start doing an agency in help out other businesses to automate their business and to be honest uh I think was a great journey for us but blue coffee it's still there okay it's still live you can find it on Blue bl. coffee you can order coffee all over Europe and and please do so awesome so and now let's dive into the use case that we have today which is still related to Blue coffee right so you actually built a chat but for this uh e-commerce website uh why did you have to do that and what does this chatbot do so you you were mentioning earlier during the kpis right because okay saving Savings in money and Savings in time are important right but sometimes uh the customer experience is even more important we had we had customers that were reaching out asking do you know where is my order do you know what is the status of my order and stuff like that and because it was manually we were replying from time to time it was on a WhatsApp sometimes P was waiting for me to reply sometime I was waiting on him to reply and when the AI um Trend started about a year and a half three years ago we were looking at how can we improve that experience can we build something with make and an AI which is connected directly to our systems and can respond directly from those systems with live information and that's why we believe uh offering like 247 support having a chat bot that's already all all the time there it's able to be connected to our system and offers recommendations about the products that was kind of the what drove us into building this product amazing yeah and I think solution what you just mentioned right now is the ability of that chatboard to check specific information about about the order and answer the user with this uh specific information so maybe we can move on to the next step which actually what do you need or what is the process of building an AI chatbot yeah so building chatbot is not something new right because there were chatbots before and also kind of powered by Ai and you can find a lot of platforms out there which can help you build like your chatbot with AI but most of them rely only on the llm training and at the same time it relies maybe on maybe you upload some documents about the business but that will not provide live uh data from your business right so so when open AI released the function capability about a year ago being able to allow the agent to think about okay I need to grab some uh order information so then I have a function that can provide that information to me right and that's when we started looking at doing this with make so how does it work we have a user is messaging the AI agent the AI agent will get the message the AI will go into a decision making what does mean it will will try to understand the question and if he already knows the uh the answer it will just reply back to the user if it needs more information which is related to uh business knowledge like if you have maybe a a vector database it will go towards that database search for that information and then go back and reply to the user and additionally we have what we call functions and those functions are um actions that are defined for the AI to take in order to get additional information or in order to perform an action towards your system and with the answer from those function he can then reply back to the user that's brilliant and like I think the important another important element here is like two parts what you have here that the the the what people are cling as rag some people know uh This Acronym some people don't so barag is is a very piece of important piece of information here and the other part is the ability to connect this bot with different application whether this going to be WhatsApp whether it's going to be uh slack or or email as well for sure so we have I think when we look from um enhancing an AI chatbot we make uh we have these functions which I call Live Connections I like to call them because in reality the uh AI agent has a direct connection towards those platforms it could be towards the CRM it could be towards an Erp it could be towards e-commerce You Name It We have what about 1,800 applications in make any of those right then we have this retrial retrieval argumented generation which we call it rag which is additional information generally organization specific knowledge that's added to a vector database where you can query it with with filters with metadata information that you can grab from this database P it towards the AI and then uh get a better answer from the AI or hopefully even a perfect answer from the AI and then the nice part again because we're using make we can connect this AI agent to any platform right if that chat system has an API behind it WhatsApp telegram slack teams hopspot any chat system with an API web hook system it goes every me every new message goes directly to make you run the AI um engine and then you get back an answer right then you respond to that uh message from the AI brilliant so that's that's we already I think we already covered the live connections we already covered the uh the rag side is there anything else you need to add uh on that site so just maybe just to mention here on live connections right you think about connecting to any system like I said you have the API functions that are powering this and any application that exist in make for that matter any application with an API because make has a great s DEC it takes a couple of hours days to create depending on the complexity a new application and I encourage all of you to do that and you can connect to any API application right and you can easily swap application and test the way you need right and you get your data into the assistant and very important don't think about just getting information inside the AI the AI could also do actions right which is again something very specific to these functions that are powered by make by allowing you to create things right think about like you want an assistant to create a deal in your CRM or create a new order for your e-commerce or create a new task in your clickup or whichever right power power actions by AI yeah yeah and yeah if you speak a little bit about the rag as well yeah rag to go a bit a bit in details uh it's a bit more complicated I really advise people who want to learn more go to YouTube search for Rag and you get a lot more information but practically uh by using make you can automatically embed data to Vector databases like pine cone it's very easy to have like a Watcher in Google Drive or a Watcher in your SharePoint any of your organization where your or organization data exist and with that you bring the uh with a watcher or with a web hook and I'll explain a bit later what the web Hook is by the way because probably some of you on the call don't know uh you instantly get the information from those platforms and you transfer it to a vector database and then the AI will have access to it also important with tools like zeroc code kit Cloud convert you can also transform the data when necessary so you might need to like OCR the an invoice or you want to convert a picture to a format that um understandable by the AI right you can use these tools to convert that information and you can send the data to any vectorized database with an API right which makes it very very uh powerful and dynamically changeable so I think maybe one easier way to explain rag is that AI model usually already trained in uh uh in a specific source of data and what you do with the rag you are adding more uh learning to thei model based on your own your own data right so like this is maybe the most simplified version of that we need to be careful here uh a lot of time like fine-tuning learning teaching the AI it's it's a bit of a different process where you need to have like a very good data which you then train the model on top so you do a fine tuning on a Model while rag what rag does you go the the AI decides it needs to query on certain topics the vector database it will get get the answer from the database and then it will feed back to the AI your question and the context of the search right so then the AI will understand hey here's the question here's the rag uh information from from the business try to reply to that question with this additional knowledge that you just got and then you get a much proper answer than without providing it a lot of time the AI initial training has no way to know internal business information your I don't know f make you frequently ask questions or contract information or offer information think about very specific organization knowledge brilliant no thank you very much all right so now let's go a little bit more practical on the actual agent that you uh that you built so how how does this looks like within your process yeah so think about the same process I described before right we have a user that's messaging the blue coffee website so the way we do that we have a a a chat which is embedded by like um JavaScript code to our website whenever a person a customer loads the website they have the chat fun uh widget they click on it and they start a conversation this chat bot is powered by voice flow voice flow it's a service that provides this kind of widgets for chat and it provides a way to create a logic into that uh for for the voice chat like different buttons for starting the conversation um and then being able to access additional data right then we have the AI agent which is inside make the make AI agent which is powered by open AI this is triggered by webbook so web Hook is um an in think about it like a link right you gets you somebody sends a message and then that data goes towards that link which is a it's it's a post of data towards that link towards that endpoint and then inside make you just get the data instantly right or in in milliseconds uh and then with that you can connect the data that you received on this web hook to the next uh steps that you need to take in the flow and we'll show that a bit live on our demo then the AI takes a decision if he needs to go to one of the functions you can see here some examples of our function is like whenever somebody starts asking about an order we want them to make sure it's the right person and they have the right access so the bot it's instruction to ask first for the email it will send a code and then verify the code so then we have a function that generates and emails the code and then we have another functions to verify the code very important here like like a note if somebody plans to do that because the the idea of this webinar is for any of you to try it yourself right and by the way if you want to get our blueprints to play with it let us know and we send out the blueprints uh very important when you generate the code don't tell the AI the code because somebody might be smart enough to quck to to keep asking questions until they get the code we just tell the AI we send a code the user will get the code and the verify code function goes to a data store in make verifies if the code is the same and then confirms to a yes the code is correct you can pass and you can give it more uh order information to that to the customer or create an order for that matter and then we have additional um functions for verify order status verify shifing status create order and so on So based on the scenarios that are powering those functions the AI will get that instant knowledge about an user order and then reply back to the user very important here the AI does not have access to all our system directly it cannot really see all the orders in our system all the data in our CRM it just gets exactly the order that the customer is asking for right and it Lees only this conversation and also maybe important for some of you on the call on the open AI API there's no training uh by open a right so all this data it's it's it has its privacy correlated with it right so we're not sharing this data with open AI in any way except with the model and they are their privacy is very strong on This brilliant that's awesome so and I think that that the next T like I mean what's what you just showing right now just before jumping into the I just want to show the people how does it look like how does these steps look like yeah so somebody messaged the assistant we have a step inside voice flow where you have the setup of the assistant you can see here whenever somebody clicks on help me with my order there's going to be a capture of the reply from the from the user and then we send that request towards this link which is the web hook which will send back the response from the AI and then it will reply to the user and then there's a loop between the a conversation Loop between the AI agent and the user or the customer on our website right and think about if you would instead of doing this inside Voice Low here if you would do that in a scenario that's starting with the WhatsApp or with the telegram um module this will be at the same time using the same agent on inside WhatsApp and helping your customer directly in WhatsApp The Next Step will be the scenario itself that gets the data with open AI module that's doing the uh computation and and going over API towards open Ai and we're getting an answer we're structuring the answer and sending it back to the webbook and then we have the additional uh scenarios which are uh function scenarios you see very easy generate a code put that in the data base reply it back uh go to same day here on the on the right top side it's our shipping provider we get the order it goes to same day by API checks if it's already shipped checks if it's waiting somewhere has the exact uh status of that shipping right now then if we generate a code you see there is an email there creating an order you see there the the Shopify by going back to Shopify to uh create the line items and then create the order for the customer awesome so I think we're going to move right now into to make to actually see how to build that yourself but also like looking at the comments on the chat uh for everyone we going to be sending you the blueprint by email so after after the webinar you will all get the blueprint uh everyone who registered to the webinar will get the blueprint for these uh scenarios and if you have questions please add them in the Q&A tab not in the chat just add them in the Q&A Tab and at the end of the webinar we'll be answering these questions Okay so let's go to let's go to make let's show this before we going into make which is very important let me just show you what are the different steps in order to achieve something like this okay so first of all you need to go to open AI website you sign up for not for chat GPT by the way you sign up for the API okay very carefully there and you go to their assistant tabs and you create an assistant right so this is where you create a new assistant in my case I have here the blue coffee bot and then you need to Define okay you define the module that's using here here I'm even using see here an an older version I could switch it to the latest one which is 40 but I'm not going to do this now maybe we're going to have an issue with it so then we have the instruction so I recommend you write as good as possible instruction to make it very clear what this AI is supposed to do right what different functionality it has and then on the button side you have bottom side you have the functions which as you can see G get shipping history get uh order create authorization code valid the user code send the generated code get order information and so on right so these are the functions that let's see like this one get order information right these are the functions that are powering this assistant it might look a bit codish and a bit complicated it's not so complicated if just read it right name description strict is it strict or not right you you just say yes yes or no and then what are the different parameters that you expect to have that you have on this give me give me just a and actually I think one one very good thing about the functions uh I think TB was saying that you can actually use chat gbt to help you write uh uh write write the function uh so uh like if it's if it's a little bit like complicated for you how can you uh how can you write such function you can actually using uh you can actually use chat gbt for that let's give TP a second just to uh go to a quieter place and again so for for everyone if you have questions please add them to to to the Q&A uh and we're going to answer them at uh at the end uh okay can you hear me right now we can let's go okay let's go so just to make it very easy for for everybody you get this code very nicely you copy it from here or for that matter you can use one of the examples like get the weather one it's very easy you copy it you want to do something think about what are the parameters that you want your function to do like if you want to like send an email you need to understand what is the email and what information do you want to send as an example right so then if you go to a CH GPT as easy as uh creating go to chat GPT say help me help me create a function to uh send an email to uh to customers based on this example and if you specify hey this is for uh open a I API and you say which par parameters and so on it will know battery to do and there is a custom GPT to do that right but I just want to show you hey it's not very hard to really do that just use CH GPT it will ask you to do that function and it will write the code for you okay try that I can okay now it's kind of because I didn't say probably it's for open AI it will not do it properly use the custom GPT for that okay then we have the voice flow setup where we have have some starting points let me close this a bit so here we have the voice L set up where I have a blue coffee bot this is this has the starting conditions right and it tells okay which are the options for the user to take you just personalize that and then you have one uh web H where you send the data and this is where we also do the tread ID so we capture here the tread ID so every time you send a new message it will go to the exact same um tread and the AI will continue the same conversation and then um there we go we go to make right so what does it happen in make so in make we have what we call a web hook so this is an instant trigger so whenever you send data to this link that we're creating now so I'm going to create a new one just say like um AI bot for example and then I get the link so all the data that will be sent to this link it will show up in this scenario all right so once we get the data from that link we go to open AI message and assistant and here okay we create we create a connection we need to select the assistant that was just built as you can see I have some here and then very very importantly all the functions that were defined inside the assistant you will see it here right so I have the different functions and then the make module is expecting us to provide a URL right so what that what that URL is if we build a function for example I have here create authorization code we have a function we start again with a web hook we have a simple generate code where we do like a generate uh random code we put that in a database and respond back right so then all I need to do is copy this address the the link from here go back to the to the scenario and put it here right so with that and then I complete all the setup of my module I'll jump into my initial scenario not to spend like 10 minutes configuring this but you can see here all the setup of the bot which is all the different function you define the message connected to the web hook you define your um somewhere here the yeah the assistant and that's it we get the message and we reply back right so then if if I go back to voice flow for a second if I didn't lose that okay so let's run it once right so I could run this directly on the website or you can run it directly here in in in um webflow in voice flow so once somebody is pushing this I need help with my order I would say hi can [Music] you give me the status of my order demos always break so I'm I'm curious if it's actually going to work never works live demo you know how that is the demo God sometimes with us the demo God sometimes without us so we'll get you see it was already triggered right to help you with your order status I will need the order reference or I can provide um something to authenticate myself right so let me just go quickly right into the into the the chat boot right here I'm going to discard you see there was a few seconds ago we had this running and this generated the answer right so you get that code generated here now if I put hey this is my email for authentication it will send me an email for that so it will start using the get function and the other uh functions to really power that okay so just to wrap up on on how this is done we have the voice flow again that's messaging the scenario uh voice flow we'll send the data to make make will get over the web hook use the open AI module to computate the answer right the AI will decide if it knows the answer for example at the beginning based on instruction will just tell you hi I'm here to support you with your order I need your ordered number right or if it needs to do something it will go do that action on those functions brilliant and I think like I mean in that in that example you were showing using open AI but like obviously you can use other uh uh AI um AI apps and make has multiple uh uh AI apps that you can use if you prefer others than uh than open Ai and then the last step that you mentioned which is the functions right so this is where you can identify what different actions you might be performing so you might be implementing that in uh another website which is not an e-commerce right so there's no there's no uh orders but you want to uh add a new a new uh contact in in yourm for example so this is up to you to configure what kind of action you want this bot to uh to perform sure I mean there there are different ways to implement this depending on the AI that you're running with uh or using because open a has this API that's managing the thread for you if you're using another API that does not have an AI that does not have a thread handling then you need to manage that yourself so you need to do to create a database with the messages feedback those messages toward the assistant right so there are different ways to implement this but yes hypothetically you could do this achieve achieve it with any AI that has an API awesome great uh and TB you said like this this agent is live right now right it's still still bringing you results bringing you uh uh Revenue hopefully right I yes it should be um and that's why I tested inside voice flow because if you go to our website my team just transition last week from prea shop to um to Shopify and when I went there today before the testing to see okay I'm going to do this demo on I saw it was not there so I need to embed the code off and that's going to take a bit few hours but we'll put actually few minutes not hours we'll put it there and you're going to test it live if you want to awesome brilliant no thank you very much um again so like uh number one we're going to be sending you the blue brints of these scenarios that uh TB was showing and uh you also going to be have the recording we going to send that to you is going to be available on the website and number three if you are not a make user yet uh you can be using the link that we will be sharing so you can use make uh two months for free uh with uh 100,000 or 20,000 operations I think right yeah and use those operation on AI testing on AI awesome so now we we're gonna move to we're g to move to questions right uh if you have any questions please add them to to the Q&A tab not to the chat add them to the Q&A tab we're going to go through them and answer as much as the time the time allows right um so the first first question I have from from Thomas how do you get keep from customers or non-customers who are looking to abuse the AI chat but I think you you uh mentioned that a little bit during uh during your presentation can you maybe elaborate a little bit I mean if if if somebody okay just speaks too much with the AI okay it will have some impact on us on cost because it's using operations and AI tokens but that that's it right uh it we are limiting it by instructions not to like provide generic information so somebody would just use it like going instead of going to the Chad GPT website uh and then from a business perspective before doing any action like creating an order or um getting information about an order we always do authentication depending on the platform could be easier or more complex like for email you generate the code and send it over if you use WhatsApp you know directly based on the WhatsApp API you know exactly the phone number you're talking with if that customer already has an account you can easily identify it if they don't have an account you will ask them how do you want to create an account right so you can uh by identifying the person you can then uh mitigate that yeah uh great thank you then the next question like someone just uh confirming about track so by using rag you are retrieving your vector base with ducks uh uh related to products recepts and uh du in general is that correct you're retrieving with the data that lives in your R Vector database it's correct I mean it could be whichever documents could be for I give you a nice example we are using internally at make it future we are using uh platform called fireflies AI which gives you like meeting notes and actions after every meeting and we have a scenario after each meeting we take that export conversation and all the information that happen on a call with our customers of course by consent with our customers we we are recording that and we're putting that into a vectory database right so then we have an internal chatbot if I'm looking to if I'm searching for a question regarding something we did internally on a meeting we find that very easily great so it could be whichever data you imagine you want do your AI to have access to yeah uh all right so still I think a lot of people are still interested in uh in the rag so there are couple of questions about buying con here and think you can share a little bit of your experience Pine con um yeah so Pine con is one of the uh vector vector data basic that is very easy to set up uh it has a very I would say um comfortable free plan we are on the free plan to be honest uh you can create a pretty big Vector database on the free plan and then uh it has a great make application works very easy with uh with the open ey model and open Ai embeddings and in my personal testing and experience it's much much better than open AI own Vector database and and that's because it it allows you to put a lot of metadata on top of the vector dat database M understood which provides you additional information about that search MH great uh someone someone's asking to explain a little bit about the functions within a open AI I think we don't have the time for that but what would you recommend them uh uh to do like where can they go to learn a little bit more about functions within open AI so open AI documentation on that is great at the same time there was a when the open I released the function about a year ago they had a presentation of about 40 minutes really showcasing that so I I encourage you go search for the openi video on the functions and then there's a ton of YouTube videos actually and I think even us on our YouTube channel or on LinkedIn on my profile you can find uh some some videos related to how we build functions make great thank you and someone someone was asking actually I saw this a couple of times about saving the answers from uh from the B so when the B is answering uh a question do you save this uh answers number one for yourself and number two do you use this answer to for future answers as well or it is like expired answer once it's sent at at this point we are not doing that uh be it could be valuable if you want to do uh fine-tuning right so you have you have a conversation you have a back and forth question and answer so then if somebody would want to do more fine-tuning on how you would expect the AI to answer by having those paing question and answers and then going over it step by step cleaning it making sure that you always like the answer uh and and put maybe positive negative answers and then you could do fine-tuning with it uh but that's a bit I would say that would be a lot more advanced a lot more time consuming for in our use case we're not saving it it just stays on the API of open AI I think their retention policy it's probably it's it's like 60 or 90 days and after that it just it's gone MH understood um then I I see like a question but also by by multiple people in different words which is how does the AI understand which function it should uh uh it should trigger the AI understands with function to trigger based on the description of the function right so if I have a function that's uh saying it's described as this function uh will give you information about an order based on the order ID right and then the parameter of that function will be let's say order ID and maybe I have the authorization code because I use that as a secondary authentication into that process right so then uh you mark those parameters as as required and the AI knows whenever somebody's asking for order information he knows he has in his pocket this function he knows he needs the order ID and the code and then it will ask ask you directly the user can you provide order ID yes okay for the code it has another function that's say hey if you want a code you need to generate a code send it to the user and you get it from the user so then it will push to the other function to generate the code and ask the AI for the sorry the customer for the code right so it's it's just you you as the developer you're describing this in real words to the AI when it needs to use that function and it the AI will trigger the function based on that explanation understood great uh another question someone is asking about like the average cost of this and of course the answer is always it depends but maybe we can just explain where where you going to need to pay like you're going to need to pay this service that service like maybe just to explain uh where the cost will be coming from I mean when when I when I really started doing this I was like I was thinking that I will use like this this is going to use like tons of operations in reality a user spends maybe five 10 messages uh on the chat because it's not like having a chatting with the a a lot he's just asking for some business for for some information about the order and it's gone right so then it's think about maybe four operations per run per message and then you have maybe 20 30 operations for the conversation and then you have the tokens on the open AI side which is again few cents I would say I I'm not depending depending on how long that answer is it might be I don't know below one cent it could be maybe after the end of the conversation three four cents but it's it's in in my in my in our experience uh for about few hundred orders that we have on the shop it it's fully justifiable understood great and someone is asking of course about uh privacy so uh are you actually sharing any sensitive uh uh info about the customers if this something that can cause concern and how are you addressing that so and that that's again that's the beauty of using make into this okay when you do the function that gets getting the information from the order you decide as the developer how much information do I send from Shopify to the AI can send everything or I can send only the status and the price the total price and and that's it if I don't want to send the customer name the shipping information the billing information I'm not sending it right so I'm just sending what I need to send or what I consider for the AI to be critical to be able to uh to answer the question understood uh and I see I see and it's just asking like yeah can you please write down the system that we are using it's a little bit difficult for her to understand uh she's Swedish none of us is native speaker anyway but anyway this is recorded you're going to get the recording you're going to see all uh the system all the systems there um I think like this questions appear before and uh maybe we can we can close with uh with this one because I know it it's quite important and I'll add it in two fault so number one again how can we avoid people trying to trick the ey into getting sensitive uh information and number two how do we avoid AI hallucination which is an always concern uh for anyone using AI so uh for your first question again you're just giving the AI the information that you consider important for for the AI to give an answer that that's the first thing so limit the information for Just what's needed uh and most of the time telling a customer hey your order is on way not business critical right or not not not an information that is critic or or the the order is waiting it's it's inside the distribution center in whichever City again not critical information uh then the open Ai and I recommend you going and check that any of you go to their privacy setting on the API and you can see they are not doing any training on that data which is very important for for us to make sure that open Ai and next open AI models will not be smarter or having information about a customer order or something right we we need to make sure that's not happening so that's how we limited and then to answer your second question about hallucination um by providing this direct information on its answer from functions the AI will have the exact information it needs to answer right by providing information from rag okay of course you need to make sure that your rag it set up correctly you're doing um a good good a you have a good strategy on rag because there's a lot of way to do rag but by providing that context information the better you can provide it the better the answer will be and I can give you I can give you a kpi on that we did I'm working on another Ai and we we've done some test testing before implementing the rec strategy directly with the open AI Vector database our accuracy for answer was around 40% so something like 30 38 39% was good answers about 30% were very bad answers hallucinations and about 30% was like incomplete answers once we move that data towards the rack system with with a good implementation right now we are about 92% accuracy wow right which is a a crazy big difference and within a week we tested six different uh rack strategies with make just because we're able to go change the way we filter to Pine Cone create the secondary scenario and actually we have six different strategies and we go to an database check the answer check the rag and we decide okay this is the best this is the one that's working the best right again because we're using no code technology it's so fast to test and deploy great awesome thank you very much TB that was very very insightful very educational I learned a lot today uh for everyone we will share the blue brands with you you will share we will share the recording of the webinar uh with you please uh do follow me we have more webinars coming BL please F uh follow tiby and reach out to him I know there are more question that we did not have the time to answer so please reach out to tiby if you have any questions he will be more than happy to uh to help you as well tiby thank you very much that was that was awesome