Enterprise AI Assistant Platform | Production-ready AI Agents Hackathon @ AI By the Bay
Recording: Enterprise AI Assistant Platform | Production-ready AI Agents Hackathon @ AI By the Bay
Can everybody hear me? Ah so my name is promote uh so what happened was like you know few days before uh I was on IBM event and I met a lady who was working on accounting firm and she was really impressed with what uh we built on that event and she sent me you know oh so that was only for David why don't we build something like a really use case in a real world scenario you know why don't you build something for me and she was saying like that sounds really interesting I'll pay you hourly you I was like why not you know and then I came across this hackathon u the hackathon judging criteria and what she wants it was kind of like you know in between it was like okay that sounds interesting maybe you know like I'll build something on hackathon and it might work for her you know like and she gave me many problem statement out of that you know because of time limit I can only uh fulfill like a few agent kind of thing which will definitely help her So yeah talking about you know the uh real world business problem. So uh I think if you guys are working on some organization where there are like 40 50 people uh let's say not too much than that because uh for that like I mean say for the organization that have lot of people there are really you know big uh companies like SAP doing all the thing but for the small companies like 40 50 people you know uh they have they might have issue they are having issue with these kind of things you know like HR assistant like the uh Like for example, if they want to find something in the organization, we can help her on her team and also you know like for example she wants to uh do like an accounting stuff like a calculation or something like that and also do like an estimate for how much do she need to pay wages for her employee you know um and then just do basic document sources. Uh that's why I so that's what brought me to this idea. So the architecture of this project is like there's an uh platform and I'm using front end react bite for the front end part and there is like a chatbot there's a orchestration chatbot here and then I've created like an AI agent which is HR agent the other one is analytics agent and the other one's document agent you can see list of all the tools and MCP I've created for this product let's go to the demo part that's interesting one all right so basically very simple platform to use. Uh in future I'm going to implement a lot of tools something like that and it's already you know ready to use that means you can just put an API to this platform and start using uh let me know before that. So at the moment I don't have enough data so I created like a few uh you know dummy data for this platform. Let's say there are like 40 people in this organization and you know like okay I want to find something like some employee I want to work with who has really good like Python skills agent agent is going through the database and finding out the people who has really good skills on the Python okay I found it oh there are like six people who have really good understanding of Python I can you know just text them and just say like oh let's work on this project okay I'm new employee let's say I'm a new employee I want to see like how they organize and chart for this firm looks like like who is you know who I wanted to take this. Oh all right I can see the C of this company is Jennifer Lee
So this is the main HR assistant kind of agent does it helps you with your HR stuff and when you have like 40 50 people or maybe 100 people in your organization it can be really useful tools for you. Uh the other one is the analytics hop. This is really cool for people who do a lot of maths on your uh you know uh on your work. Uh for example, I want to play calculate like how much do I need to pay for an employee with the federal tax rate, state tax rate of this one. So you can use quick big quick book sorry but it this will help you give an estimate of you know like just in like one minute it will give you estimate of like how much uh do I need to pay my employee and how much do I need to pay for tax and it's not only about that you know like she told me like sometime you need to for doing a fee proposal or something like that you need to do everything manually on Excel or something like that so she said why don't we implement those technology on you know this agent platform which can do maths for you like just to calculate like free proposal do a free proposal for you and yeah so basically this agent have that capability so it can do the maths really well so I've implemented the agent really good tools for this agent uh it's taking time [laughter] uh please all right yeah it gives you the calculation for that one and then yes so you can see the calcul I'm using chain of thought reasoning prompt for this one and The final was one one is the document intelligence. So it's basically rack framework which helps you to understand the uh you know basic policy of the companies and all those things. So the internet is really slow uh that's why you're getting that it's not only about building right it's also about evaluating like how your product is working is it working fine. So for that one I have implement the opic comment for that one
So you can I can see the every tra traces of you know like what they talk about you know like I mean say the chat with the LLM kind of thing and I can also do the evaluation for this one. So like how good the platform is working. So I've used the uh LLM is just for evaluating my uh platform. So I can see here you know like okay so I can see the list for this response you getting minus one because it didn't answer that question properly. So, so that's how I am evaluating this platform and talking about you know the uh security I implement the guardrail AWS bedrock guardrail for this platform which means like you it will help to avoid the harmful content for the agent. So basically it's like a proper agent. It's not just agent it's like a proper agent that can help your business work really well. So what's next? The next thing I'm going to implement is the age framework
So few days before there was like a um yeah like a like a small talk by the Vincent from the co like OPEC comment he was saying like the prompt is dead prompt engineering those are just useless and he was right like you know like when the context of the you know uh context for the agent increases with time the performance decreases so what I'm working on here now is the agent context engineering kind of framework so this will be implemented soon so because of time limitation I couldn't do this one. So basically it does what it does is like uh there are like three different agent and it has a strategies to reduce content just and evolve uh with time. So that's what my agent does. Thank you for your time. [applause] Any questions? Any questions or something you want to see? Nothing. Okay. Thank you. Thank you.