rag-fusion and OpenRag

RAG-Fusion's multi-query generation and reciprocal rank fusion technique represents a retrieval strategy that could be integrated as one component within OpenRag's broader multi-strategy architecture, making them complements rather than competitors.

rag-fusion
66
Established
OpenRag
40
Emerging
Maintenance 20/25
Adoption 10/25
Maturity 16/25
Community 20/25
Maintenance 10/25
Adoption 7/25
Maturity 11/25
Community 12/25
Stars: 908
Forks: 110
Downloads:
Commits (30d): 7
Language: Python
License: MIT
Stars: 36
Forks: 5
Downloads:
Commits (30d): 0
Language: Python
License: MIT
No Package No Dependents
No Package No Dependents

About rag-fusion

Raudaschl/rag-fusion

RAG-Fusion: multi-query generation + Reciprocal Rank Fusion for better retrieval-augmented generation. Includes evaluation harness with NFCorpus/BEIR.

Multi-query generation is powered by OpenAI's GPT to explore different facets of user intent, with results aggregated via Reciprocal Rank Fusion for re-ranking across multiple retrieval perspectives. The implementation combines vector search (ChromaDB) with optional BM25 keyword search in a hybrid architecture, and includes a quantitative evaluation harness against NFCorpus/BEIR benchmarks showing +22% NDCG and +40% recall gains over baseline vector search through diverse prompt variants and result fusion strategies.

About OpenRag

incidentfox/OpenRag

Multi-strategy RAG system achieving 74% Recall@10 on MultiHop-RAG. Combines RAPTOR hierarchical retrieval, knowledge graphs, HyDE, BM25, and Cohere neural reranking.

Implements a FastAPI server with pluggable retrieval strategies (semantic search, HyDE query expansion, BM25 hybrid matching, multi-hop decomposition) that run in parallel before Cohere neural reranking, with built-in persistence for RAPTOR hierarchies and a comprehensive benchmark suite supporting MultiHop-RAG and CRAG datasets. Ablation studies show the neural reranker alone contributes +9.3% recall improvement, while local cross-encoder alternatives are available for privacy-sensitive deployments.

Scores updated daily from GitHub, PyPI, and npm data. How scores work