Personal Project

Vita AI started as a broader product idea around an AI-powered life assistant, but one of the most serious technical parts became the RAG evaluation playground built on top of a structured product documentation corpus. Instead of making just another chatbot demo, the goal here was to understand how different retrieval systems behave on real product-support and product-knowledge questions.

The project focuses on comparing Traditional RAG, Vectorless RAG, and Agentic RAG across the same documentation base. It is designed more like an applied AI systems project than a UI-only app, with attention to ingestion, chunking, retrieval quality, benchmarking, failure analysis, and evaluation logic.

• Built a structured documentation ingestion pipeline for product knowledge and support content
• Designed the project to compare Traditional RAG, Vectorless RAG, and Agentic RAG in one playground
• Worked on chunking, retrieval strategy, evaluation thinking, and benchmark-oriented AI system design
• Used a product-like stack with Node.js, TypeScript, PostgreSQL, Prisma, and Firebase in the broader Vita AI setup
• Focused on making the project serious enough for portfolio and research-style discussion, not just a chatbot demo

RAG Playground Overview

Main interface showing the RAG playground flow, system variants, and evaluation-oriented structure.

Documentation & Retrieval View

Structured product documentation view used as the base corpus for retrieval and benchmarking.