TT./projects/hybrid-rag
GitHub ↗

Hybrid RAG System v1.0.0

Dense + Sparse + Graph retrieval · Corrective RAG via LangGraph · Feedback-driven reward model · RAGAS evaluation

PythonFastAPILangGraphQdrantElasticsearchNeo4jReactDocker

Hybrid RAG system

Ask a question about ingested research papers

Try: "What evaluation metrics does the attention paper use?"

Pipeline status
Qdrant (dense)
Elasticsearch (BM25)
Neo4j (graph)
PostgreSQL
Retrieval config
Embedding: all-MiniLM-L6-v2
Chunk size: 512
RRF k: 60
CRAG threshold: 60%
Max retries: 2
Reward blend: 70/30
Indexed corpus
12 papers
2,847 chunks
423 entities
1,156 relationships