How fan-luo/legal-rag works
Legal-RAG serves statutory-question answering through an API: requests are routed, retrieved from language-specific legal corpora, and answered with provider-agnostic generation. Its other major runtime capability is PDF ingestion with background index updates. The README documents offline corpus/index construction and hybrid retrieval; build scripts are omitted from this runtime map. Connections are limited to README statements and source-index references or excerpts, so some internal calls are not shown.
Main entry points: hybrid_retriever.py Hybrid retrieval · server.py HTTP API · orchestrator.py Index updater · rag_pipeline.py RAG pipeline
Components 27
Online serving
- HTTP API
legalrag/api/server.py - RAG pipeline
legalrag/pipeline/rag_pipeline.py
Legal retrieval
- Language routing
legalrag/retrieval/by_lang_retriever.py - Hybrid retrieval
legalrag/retrieval/hybrid_retriever.py - Dense search
legalrag/retrieval/dense_retriever.py - BM25 search
legalrag/retrieval/bm25_retriever.py - ColBERT search
legalrag/retrieval/colbert_retriever.py - Result reranking
legalrag/retrieval/rerankers.py - Graph expansion
legalrag/retrieval/graph_retriever.py
Routing and generation
- Query router
legalrag/routing/router.py - Issue extraction
legalrag/routing/legal_issue_extractor.py - LLM client
legalrag/llm/client.py - LLM gateway
legalrag/llm/gateway.py
PDF ingestion
- Ingestion service
legalrag/ingest/service.py - PDF parser
legalrag/pdf/parser.py - Chunk and validate
legalrag/ingest/ingestor.py - Background queue
legalrag/ingest/task_queue.py - Index updater
legalrag/ingest/orchestrator.py
Index and graph
- Vector index
legalrag/retrieval/vector_store.py - Law graph
legalrag/retrieval/graph_store.py - Incremental dense builder
legalrag/retrieval/builders/incremental_dense_builder.py - BM25 builder
legalrag/retrieval/builders/incremental_bm25_builder.py - ColBERT builder
legalrag/retrieval/builders/colbert_builder.py - Graph builder
legalrag/retrieval/builders/graph_builder.py
Other components
- Questioner
- Model provider
- PDF uploader
Connections 29
- Questioner → HTTP API: submits query
README.md - HTTP API → RAG pipeline: invokes RAG
legalrag/api/server.py - RAG pipeline → Query router: routes query
legalrag/pipeline/rag_pipeline.py - Query router → Issue extraction: extracts issue
legalrag/routing/router.py - RAG pipeline → Language routing: retrieves by language
legalrag/pipeline/rag_pipeline.py - Language routing → Hybrid retrieval: dispatches search
legalrag/retrieval/by_lang_retriever.py - Hybrid retrieval → Dense search: runs dense search
legalrag/retrieval/hybrid_retriever.py - Hybrid retrieval → BM25 search: runs BM25
legalrag/retrieval/hybrid_retriever.py - Hybrid retrieval → ColBERT search: runs ColBERT
legalrag/retrieval/hybrid_retriever.py - Hybrid retrieval → Graph expansion: runs graph search
legalrag/retrieval/hybrid_retriever.py - Hybrid retrieval → Result reranking: reranks candidates
legalrag/retrieval/hybrid_retriever.py - Dense search → Vector index: queries vectors
legalrag/retrieval/dense_retriever.py - Graph expansion → Law graph: expands graph
legalrag/retrieval/graph_retriever.py - RAG pipeline → LLM client: generates answer
legalrag/pipeline/rag_pipeline.py - HTTP API → LLM gateway: configures provider
legalrag/api/server.py - LLM gateway → LLM client: wraps client
legalrag/llm/gateway.py - LLM client → Model provider: calls model
README.md - PDF uploader → HTTP API: uploads PDF
README.md - HTTP API → Ingestion service: starts ingestion
legalrag/api/server.py - Ingestion service → Chunk and validate: parses document
legalrag/ingest/service.py - Chunk and validate → PDF parser: extracts PDF text
legalrag/ingest/ingestor.py - Ingestion service → Background queue: queues work
legalrag/ingest/service.py - Ingestion service → Index updater: starts index updates
legalrag/ingest/service.py - Index updater → Incremental dense builder: adds dense records
legalrag/ingest/orchestrator.py - Index updater → BM25 builder: updates BM25
legalrag/ingest/orchestrator.py - Index updater → ColBERT builder: rebuilds ColBERT
legalrag/ingest/orchestrator.py - Index updater → Graph builder: rebuilds graph
legalrag/ingest/orchestrator.py - Incremental dense builder → Vector index: updates vectors
legalrag/retrieval/builders/incremental_dense_builder.py - HTTP API → Questioner: returns answer
README.md