AI/ML
G4 Expert System — Session-Isolated RAG Chatbot
Lead developer (team of 5): RAG chatbot with ChromaDB session isolation and hybrid Ollama/Gemini routing, deployed on a VPS.
Python FastAPI React ChromaDB Ollama REST API
Problem
The class requirement was a basic Gemini-API RAG demo. I wanted to feel industry-level AI architecture: multi-user isolation, local-vs-cloud trade-offs, and real deployment.
Solution
Built beyond the requirement — sole code contributor (30+ commits): per-session isolation via ChromaDB metadata filtering, user-controlled hybrid routing between local Ollama and Gemini 2.5 Flash (BYOK, keys never stored server-side), dual 768-dim embeddings, deployed on an OpenCloudOS VPS via systemd + Cloudflare Tunnel.
Results
- Completed Expert System final project, semester II 2025/2026
- Fixed 10+ retrieval bugs (embedding mismatches, null injections, leaks) in a day
- Live deployment serving real sessions