Hybrid retrieval + evaluation harness for enterprise knowledge bases — 设计文档（spec-only），暂无代码
本仓库目前只有这份 README——没有任何代码。 下面的架构图、接口与指标是设计目标，不是已实现功能。
- 相关实现进度见姊妹仓
enterprise-rag-saas（同样处于未完成状态，见其 README）。 - 如果你在找现在就能跑的 RAG 方案，请使用成熟开源：LlamaIndex、LangChain Retrieval、txtai 等。
A production-oriented RAG (Retrieval-Augmented Generation) system designed for enterprise deployment, with built-in evaluation and hybrid retrieval strategies.
- Hybrid Retrieval: Combines keyword search (BM25) and vector search (ChromaDB) for maximum recall
- Evaluation Harness: Built-in retrieval accuracy + answer quality scoring — don't just build RAG, prove it works
- Multi-tenant Architecture: Designed for enterprise deployment with tenant isolation
- Chunking Strategies: Configurable chunk size, overlap, and semantic chunking options
- Source Attribution: Every answer links back to source documents with relevance scores
- Cost Tracking: Token usage and API cost monitoring per query
┌──────────────┐ ┌───────────────┐ ┌──────────────┐
│ Query │───→│ Hybrid │───→│ LLM │───→ Answer + Sources
│ (user) │ │ Retriever │ │ Generator │
└──────────────┘ │ (BM25+vector)│ └──────────────┘
└───────┬───────┘
│
┌───────┴───────┐
│ ChromaDB │
│ (vectors) │
└───────────────┘
# Index documents
python -m enterprise_rag index --source ./docs/ --chunk-size 512
# Query
python -m enterprise_rag query "What is the vacation policy?"
# Run evaluation suite
python -m enterprise_rag eval --test-set eval.jsonl| Metric | What It Measures |
|---|---|
| Retrieval Recall@K | Did the right chunk appear in top-K results? |
| Context Precision | How many retrieved chunks are actually relevant? |
| Answer Faithfulness | Does the answer stay grounded in retrieved context? |
| Answer Relevance | Does the answer actually address the question? |
| Cost per Query | Token usage × pricing |
Design / spec only. 仓库当前不含实现代码；实现推进时更新本节。诚实状态优于虚假徽章。
MIT