DocMind
Ask any PDF a question and get an answer you can check.
A production-grade RAG pipeline focused on retrieval quality and verifiability. Every answer cites the exact source chunk it came from, so you never have to take the model's word for it.
- Node.js
- TypeScript
- OpenAI API
- pgvector
- PostgreSQL
- Prisma
- Docker
- 01
Built a full ingest pipeline: PDF parsing, 500-token chunking with overlap, batched embeddings, and vector storage in pgvector, so any document is queryable within seconds of upload.
- 02
Implemented cosine similarity retrieval on pgvector's native operator, with optional HNSW indexing for scale and per-document scoping for multi-tenant use.
- 03
Engineered a streaming generation layer with citation parsing: the model references numbered sources that map back to real chunks and render as clickable references in the UI.
- 04
Wrote a Precision@3 evaluation script that scores retrieval against a test suite, measuring whether the right chunk lands in the top three and logging citation rate across every question.