patelpattu90-ai/hybrid-rag-knowledge-engine
Production-style Hybrid RAG system with semantic + keyword retrieval, re-ranking, and evaluation framework.
What's novel
Production-style Hybrid RAG system with semantic + keyword retrieval, re-ranking, and evaluation framework.
Code Analysis
22 files read · 5 roundsA Gradio-based RAG chatbot that retrieves answers from FastAPI documentation using hybrid search (FAISS + BM25), cross-encoder reranking, and LLM generation via Groq, with multiple defensive layers (guardrails, PII scrubbing, caching, query rewriting, tool routing, quality scoring, and compression).
Strengths
Well-structured pipeline with thoughtful defensive layers (two-tier PII scrubbing, 3-layer guardrails, safe fallbacks everywhere), clean dataclass-based result containers, and good separation of concerns across modules. The compressor and doc quality scorer show genuine engineering thought with proper fallback logic.
Weaknesses
The README claims 'RRF fusion' but the code uses a simple weighted average; main.py imports a non-existent function (ingest_urls); ingestion has massive URL duplication; the system makes 10+ LLM calls per query contradicting the 'under 3 seconds' claim; tests only cover regex-based layers with no integration or pipeline tests; the evaluator uses naive substring matching for relevance.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
25
Contributors
1
Files
41
Active weeks
2
Repository
Language
Python
Stars
1
Forks
0
License
—