Featured project
Local RAG for PDF Documents
A fully local RAG pipeline for PDFs: PyMuPDF extraction, Ollama embeddings, dense/hybrid retrieval with Cross Encoder reranking, and grounded answers whose evidence is extracted verbatim by a Python resolver, not invented by the LLM. Includes a benchmark-tuned abstention gate, a versioned evaluation harness with honest per-area metrics, and a web dashboard.
- Python
- Ollama
- Hybrid Search
- Cross-Encoder Reranking
- FastAPI