LEADisDEAD/Vector-Forge
Production-style Retrieval-Augmented Generation (RAG) system with FAISS-based semantic search, intent-aware retrieval, and fully local LLM inference.
What's novel
Production-style Retrieval-Augmented Generation (RAG) system with FAISS-based semantic search, intent-aware retrieval, and fully local LLM inference.
Code Analysis
9 files read · 3 roundsA Flask-based RAG system that performs hybrid dense+sparse retrieval over uploaded documents, reranks with a cross-encoder, and generates citation-grounded answers via a local or API LLM.
Strengths
Clean layered architecture with genuine two-stage retrieval (FAISS IVF + BM25 fusion + cross-encoder reranking), well-structured prompt engineering for citation grounding, and dual LLM backend support. The code is readable, modular, and the retrieval pipeline is correctly implemented.
Weaknesses
No persistence (all in-memory), no real test suite (only a hardcoded evaluation script), critical nlist=1 bug defeats IVF indexing, unbounded answer cache, and no error handling around LLM calls or file operations. The 'production-style' README claim overstates the system's robustness.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
22
Contributors
1
Files
14
Active weeks
3
Repository
Language
Python
Stars
1
Forks
0
License
—