End-to-End Agentic RAG
A local agentic RAG pipeline combining FAISS dense vectors and BM25 lexical search with an iterative tool-using loop that verifies each drafted claim against retrieved chunks before emitting inline citations.
- Period
- October 2026
- Release
- Public source project
- Work type
- Independent open-source project
- Development
- AI-assisted, with verification
What I built
A local agentic RAG pipeline combining FAISS dense vectors and BM25 lexical search with an iterative tool-using loop that verifies each drafted claim against retrieved chunks before emitting inline citations.
Developed with AI coding assistance. Source code, reproduction instructions and tests are available in the linked repository.
What was checked
7 automated pytest checks passed locally on 2 November 2026.
Verification date: 2 November 2026. Tests cover the included fixtures and do not establish production performance.
What this demonstrates
Offline tests use a deterministic hashing embedder and mocked LLM responses on sample documents. Retrieval precision and citation quality in live runs depend on the external model and corpus.
All projects