ZenMind

polished prototype

RAG / Retrieval

An Obsidian-vault assistant built around a simple rule: do not bluff when the notes are thin. It streams citations, checks confidence, and runs a corrective HyDE retrieval path when the first answer looks weak.

FastAPILangGraphpgvectorbge-m3BM25+RRF

Architecture

A multi-node graph (reformulate → retrieve → assess confidence → answer → fallback) over ChromaDB or Postgres/pgvector, with bge-m3 embeddings, runtime knowledge-base switching, a swappable embedding model with re-indexing, FastAPI + LangGraph, React 19 + Vite, and mypy strict.

Talk about work like this →