AI-Powered Chat System (Ycotek)In progress

Problem

A retail conversational AI needs to hold a coherent, useful conversation with a customer. That means reasoning about what the customer is actually asking, not just returning relevant-looking text.

Solution

A chat system with a Next.js frontend and a Python backend. My work has spanned both sides: building the interface and product-facing website, and backend API/architecture work. Working alongside data engineers, I've helped integrate LangChain and LangGraph so the system can move from a single-pass response pipeline toward agent-based workflows.

My Contribution

Frontend and backend development, API design, code reviews, and mentoring of junior developers, plus hands-on work integrating LangChain and LangGraph with the data engineering team.

Technical Details

• –Next.js / TypeScript frontend
• –Python backend and API layer
• –LangChain and LangGraph for agent orchestration
• –Moving toward multi-agent, tool-using workflows

Challenges

This is where my thinking shifted the most. With a conventional web application there's usually a specific line of logic you can trace back to and fix. With an AI system, a model can have the right information and still answer incorrectly, or an agent can have the right tool available and still call the wrong one. Simply adding more context doesn't reliably fix either failure mode. That gap is what pushed my interest from 'get this into production' toward 'understand why it behaves this way.'

Results

The system is live and in active development. I can't share internal metrics or architecture specifics since it's a proprietary employer product, but this work is the direct source of my interest in RAG, knowledge representation, and agentic systems.