welcome everyone and thank you for joining this presentation today uh I'm very exciting to be here today and I see so many people so it's great to to be here so quickly uh first let me introduce myself and my company so I'm Jeremy I am the co-founder and VP of engineering at New Logic so New Logic is a software consultancy that we funded in 2016 and we mainly build Innovative software Sol tion we are also uh Odo partner since 2021 so a bit more recent and I'm proud to say that we are over 75 people now we are located in five different countries so all over southeast Asia I I am personally based in Bangkok Thailand we have teams in uh Singapore the Philippines Vietnam and Laos so today we are going to talk about AI so AI has become very popular uh and this mainly thanks to chbt I'm I'm pretty sure all of you have tried at least chbt and it's a game changer so but first I I wanted to give you a bit more um information about what AI is and what specifically a chbt is doing with AI so this type of AI is called generative AI generative AI means that it's an AI that's going to be able to create new content so new content such as text images but even videos are audio files so and more specifically uh chpt used one uh model to generate text that is called llm llm stands for large language model and as you might know llm is based on and it based and trained actually on a very large data set so basically everything that's on internet got into that model being trained and here you you get something like Char GPT but what what's Al interesting with llms is that you can augment them so more than training them you can augment them with a contextual data and if you look at ODU we get a lot of data in ODU right and you get your company's data so if you think of this data as contextual data and you combine it with an AI you get something very powerful so this was ex exactly our ID uh on one side we get Odo with many different apps uh holding all your data and on the other side we now get a different sort of AI so we have open AI obviously but we have also some Alternatives and some of them are open source you combine the the the the two and you get very powerful tool so our our first idea was like this is very nice what can we do with that and luckily in Odo we have an app called discuss I'm sure you all know it this is where you can talk with your colleagues and so our first idea was to integrate a chatbot into uh ODU thanks to this so that's what I'm going to uh demonstrate today so here we are so we are in ODU and if we go to discuss as you can see there is a new uh button that we added it's called start AI chat if you click on it very similar to uh what exist in ch you get a new AI session and from there you will be a to actually communicate with the AI and what we did is that we plugged thei on the user you might know in ODU which is called ODOT so in that case uh you type a message ubot and uh it's actually answering but behind it it's the AI model that's answering and as you can see here the ODOT is uh responding with my name so that's quite interesting right and let me ask um like how welome how do you know me and as you will see yes of course dii knows me because well it do knows my name my role and many other details about me and that's because we have integrated the uh ODOT so the AI with the data that's provided by ODU so the the a has access to the employee database knows me my role who I am working with in what team and so on so in that way the AI is able to respond in a much better way um on top of that you might know on CH gbt you are limited to um the data uh for which the model has been trained with and there is a certain point in time where it stopped and everything that newer than that is not available to the AI bot so what we did is that we connected our AI with internet so let me ask a question for example at what talk at at what time is my talk to that Theo experience and as you will see this information of course is not part of the train data but it could get it directly from uh from internet so we see that we got access to a nice chatboard uh and the chatbot knows who I am has access to Internet now what I'm going to show is that actually we already also connected other applications with the chatbot so for instance uh in my company we have our uh policies part of Odo knowledge so it's very nice way to share uh information uh within the company and this is where we have our policy right and so what we did is that we connected the AI with Odo knowledge so everything that's part of Odo knowledge can be retrieved by the AI and can be provided to the user so let me show you one example um yeah so today my presentation here at experience I would like to know know how I should dress up because I know we have a dress code policy in our database so ODOT is typing and here it is so this piece of information was directly taken from Odo knowledge so whatever you have uh an AUD knowledge within your company this can be used as context data for the AI let me show you a second example uh because apparently I have to wear a shirt for my present uh so I'm asking what's the um expense claim process at my company and here it is so I get a detailed list of all the different steps uh which we need which we require at New Logic for uh for claiming and expense so that's very interesting and it's very very powerful it's just like few examples but there is much more you can do definitely and and you can really leverage Odo knowledge in a way that's going to um you know allowing the uh AI to to to respond in a very good way um on top of that uh we thought it would be uh interesting to her uh get access to oops sorry to different type of profiles so here I was just talking with one standard AI but you can actually what we did is that you can actually Define your very own AI by by that I mean profile so we call them experts so here we have four different experts so we have a copywriter a self speci specialist senior developer project manager I'm going to show you an example with a with a senior developer this AI will be spe specialized in uh development right so we'll be able to respond with much accurate answers when it comes to uh development of programming and on top of that so it's available in discuss so what you you can also do is include some of your colleagues in a group chat and all of you will have access to this AI so let me show you uh here I include one of my colleague called Jor Dev here it is I'm going to ask a question to uh ODOT so uh basically a simple python L World script that should be easy uh for a senior developer right ODOT is uh responding with u a simple uh python script and now you see my colleague Junior Dev can interact as well with the AI within this group chat so my colleague is asking some more details on um a specific part of this code and here it is the ODOT is able to to respond to that so it really offers uh many uh possibilities and this is just an example so to summarize what you get with this chatbot is basically everything you would expect from something like chpt but on top of that it's integrated into oo so directly into discuss and it's available to all your employees so as you know with chbt you have to have all your own account uh if you want to implement it in your company that might be a bit tricky uh here you get access to it for all your employees and uh on top of that it's based on your company's very own data so the the data is taken from knowledge emplo employee CRM project and so on there is basically No Limit wherever you have data in ODU it can be integrated within this uh AI so one question you might have uh is about permissions uh because you don't want your uh AI chatbot to uh reveal some information that someone is not supposed to see right so so obviously what we do is that we enforce the permissions of ODU so thei is never going to be able to access anything that the requesting user shouldn't have access to uh we also have experts so experts are here to provide specialized assistance you may need some very specific skills as that can be configured uh for the AI but on top of that if you think of having your experts defin in noo that means you provide them to your employees right right so in that case you can make sure that if you want a very specific tone when it comes to writing content you can Define that and all your employees will um you know inherit from this tone uh while they use the the chatbot as well as defining some company rules like if we are going back to uh programming it could be like some uh very specific code style and so on it's available in group chat as I as I showed and it works in any language so basically if you have some employees being more comfortable with a different language and you have your company policy in English for instance they can ask their question in their own language and get the answer on their with their own language but still based on the company data which is for instance in English so that was our first integration the chatbot uh it offers many many possibilities but we went a bit further with what we call the AI toolbar so I'm going going to uh demo that as well so this time let's take for example the CRM application and let's say you have one opportunity in your CRM application you got a meeting with your customer let's say last week you took some notes unfortunately the notes are on a piece of paper so you scanned it put it in this opportunity and now I'm going to show you that with the AI toolbar you can actually hear in the internal notes you can write some instruction and here we're asking uh the AI to actually transcript the text on this picture into uh text uh so we write the instruction we select the picture plus the instruction here you see we have a new AI functionality within the toolbar and I can simply apply instruction and as you will see uh the is doing the hard work and here it is we have our the content of the pictures right the notes in text format so this is very powerful uh it's using the latest uh open AI API uh the vision API super powerful and like there is so many things you can do with that but that's uh there is more uh let's say you select uh the uh the list of uh bullet points uh within your notes go going back to the AI toolbar we can uh click on this option which is continue writing so it's actually going to expand your notes with some more details as you can see instead of having a list of bullet points we get three paragraphs um here it's the list of requirements and deliverable so now I would like uh remember the experts now I would like one uh project manager to actually break this down into uh some phases and uh task so going back to my AI toolbar expert I choose my project manager and we have a task breakdown um option here so it's working on it and here it is for our different phases we have a clear list of uh task uh that should be done for this project so what I showed obviously um you select your text you apply instructions uh it's nice and it's even better than that in a way that uh it's actually using the data of this opportunity so everything with in this opportunity is used and provided to the AI to be able to respond with the best uh content because just before I was just selecting a list of bullet points right there there is no way that thei is able to construct something interesting just with those bullet points it's actually taking the whole content of this opportunity including the customer name the timeline the budget and everything and remember that di knows me my company and all of its data so it has all the information to really answer in a good create content in a good way so last thing I wanted to show you on this is that instead of just selecting text exist existing text I can also write some um instruction us AI toolbar and here I would like to have the potential uh next steps uh for this opportunity and here it is we have a couple of of options when it comes to what to do with this uh customer this is available everywhere you can write text in ODU uh meaning that here we are showing it in uh the CRM but if we go to send message I would like to send an email to my customer we have a text box so we can use the AI toolbar so here I am asking di to write an email to my customer and remember it's going to use the content of the opportunity so all the data of the opport opportunity will be used to actually create the content of this email so we apply instruction and we should get a nice email to my customer here it is so it gets like everything in it I can send it review it send it and then all right so this is uh obviously like very very uh powerful uh as I said it's available everywhere in noo we developed a couple of predefined options such as rephrase expand summarize translate and so on but you can as well write your own instructions as I said it's aware of the context so everything that generated is based on your existing data it has access to Internet and as you um as we we we saw together uh it can perform OCR so it's work in progress because the API is very recent so let me uh give you some more details on our architecture because I showed two integration but this is possible because we developed one uh what we call our neic AI uh in ODU so what we do is that so in od we have different applications uh holding your data what we do is that for each of these application we take the data model and we create either a text or vector representation of this data meaning that is going to be data that could that can be understandable understand by thei when it comes to Text data it's used as contextual data and when it come to Vector data it's used uh within a vector database so it's basically data that can be used by the AI so when a user uh request something we have our agent Computing this data together with the different AIS so we have we are AI agnostic so what I showed is based on open AI but we can work with other type of AI then this is going to create a response uh either to create new content or to interact with uh the chatbot our last integration we call it smart action and as we saw there is many possibilities and here today I'm going to show you one uh Smart Action so going back to ODU uh going back to uh discuss this time and to my uh previous uh AI session you know I am here with you today but I'm based in Thailand I'll go back I travel back to Thailand in a couple of days and I will need some uh days off to days off to recover from my jet lag so let me ask the ODOT to actually uh create this time of request on my behalf so this is what I just asked and as you can see the ODOT responded and my request is made I can just uh basically open it review it and if everything is fine you see that the days I requested for and I have my uh time of request uh done for me just have to ask ODOT for it I think that's pretty much it um even though there is one more thing uh I wanted to to to show to you today it's not only about text uh generation uh we can actually also create and generate image so for that we are using daddy 2 open I just announced the availability of daddy 3 so we have to switch to that but as you can see here I'm requesting a thank you picture basically for you attending my presentation today and that was generated by so thank you all thank you so much so if you have any question uh there is a QR code with a pad available otherwise feel free to reach out to us and we will be here the the the three days of the event so it will be very uh good to have your feedback on that thank you there are actually some questions if you'd like to Ste a little longer um a main point was about the um data access restriction um how do you manage that and could a user via the AI have access to say salary information yes yes for sure so what we do as I showed in the architecture slide what we do is that for each piece of data in ODU we have this uh Vector representation and then it's just following the same permission as oo so it's just like one additional field on the model which is Vector representation and we enforce the permission like that thank you um oops about um how exactly does the AI have access to the say employe uh employee database sorry do you pretend uh prepend context in the prompt or is there an SQL agent okay good question guys please yeah so um we we got the two different type of uh data access for sometime we use contextual data so it's basically data data that is put within the prompt but as you know we are limited in the quantity of data we can have in a prompt so on top of that for instance with Odo knowledge can be lots of data so we uh for that type of data we use a vector database so it's within like a request before actually asking uh the AI to respond all right thank you um can you give feedback about the response you got from the AI to improve it for future answers or even uh make sure that no false answers are given three sorry three what uh can can you uh give feedback to the answers that are given to improve it yes definitely So within uh within discuss um actually we can I can go back to discuss as you can see yeah there is uh helpful yes or no and this is going to be used uh to um tell the AI that the response was actually useful or not and you can even like delete the message it's not going to be taken into account uh for the rest of the conversation um could it respond to standard help desk tickets about configuring ODU for example definitely uh there is many many more possibilities we just shown like few places where we did the integration but help help desk is another ID the live bot within the e-commerce website is Al one and knowing that your that your chatbot could have access to your product it will be like very well uh informed when it comes to responding to your potential customers all right um can you share chats with other colleagues share chats with other colleagues yes yes you can uh so you can do group chat and that's going to share whatever you you said with the chatbot all right um what about data protection no one would want it company data to be accessed and stored by yes very very good point very important question so we are using open a apis So within well open a is saying that your data is not going to be used for training the AI when you use the API you have like a specific contract with open Ai and is not going to be used it's different on chbt uh chbt free version whatever you uh write in there it will be used for training jgpt but on top of that as I said we are AI agnostic so if you would like to have and use an open source uh AI model on premise in your company that's Al possible all right um are the prompt free or is there a limit in usage the prompts every time I guess it um so there is a limit in the size of the of the prompt but we make we have a specific mechanism where we make sure that if it's too long we summarize it so it's always going to work all right um could you use voice to have it work under road map um all right um someone that has its company working with external clients how do you integrate the chatbot with external clients so it's not something we have done yet but we would have to check the specific type of user to make sure we are not um leaking uh company's data to external uh users all right maybe one last one um is the app available on the app store or will it be no so it's not available on the App Store because well first we we developed it because we think it's it's a good idea and uh we use it at New Logic basically and but we think that to have to really benefit for for from it we need like a custom integration so this is why it's not in the App Store but we pos position ourself more as a partner and if that's something you're interested in I think it would be best to do a custom integration with your specific needs all right I think that's it thank you very much all right thank you all for joining thank you so much