Project Description

This is an AI-powered product research tool built for a school project, exploring the use of emerging technologies like Retrieval-Augmented Generation (RAG) and AI Agents. The core idea is simple: You input an image of a product, and the tool runs a pipeline that collects reviews and feedback for that product and its competitors. The collected data is then processed into meaningful, actionable insights. Agents are employed to validate and extract relevant information from official product manuals, which are embedded into a vector database. These manuals, combined with user feedback, are used in a RAG system to generate targeted design improvement suggestions — for example, specific changes like "Increase maximum table height from 70 cm to 76 cm." The code was designed to execute end-to-end from app.py. While it currently functions as a proof-of-concept prototype, some bugs remain, and full flawless execution may require additional patches.

Project Showcase

My Reflections & Learnings

This project began as a school assignment meant to introduce APIs and LLMs through a basic AI review system. However, I saw it as an opportunity to go beyond the scope and apply my experience in AI by building a full agentic RAG pipeline for product research. At the time, the agent ecosystem was still fragmented—LangGraph, LangChain, and OpenAI agents were all evolving rapidly, and the Model Context Protocol (MCP) had not yet been released. Midway through, Google ADK was launched. I explored it, reverse-engineered the documentation examples, and successfully got a mini test case running. However, integrating ADK into my existing workflow caused instability, so I reverted to LangChain agents—the framework I initially used—despite its upcoming deprecation. Overall, this project allowed me to apply and stretch my skills in real-world agentic system design, while adapting to a fast-moving development landscape.