How imprintlab/medical-graph-rag works
Medical-Graph-RAG turns medical documents into a searchable graph and uses retrieval to answer a clinical question; the README also describes hierarchical graph linking and a simpler vector-RAG baseline. The graph highlights the documented build and inference stages, plus nano_graphrag’s internal graph, storage, and LLM layers. The README establishes the product workflow; source excerpts and the SOURCE INDEX support the internal nano_graphrag and selected CAMEL relationships. Many listed CAMEL modules are framework capabilities, and their runtime wiring into this application is not shown.
Main entry points: run.py Run Workflow · graphrag.py GraphRAG Runtime · retrieve.py Medical Retrieval · manager_node.py Workforce Manager
Components 27
Medical RAG
- Run Workflow
run.py - Medical Data Load
dataloader.py - Document Chunking
data_chunk.py - Agentic Chunking
agentic_chunker.py - Graph Construction
creat_graph.py - Graph Cleanup
cleangraph.py - Graph Linking
three_layer_import.py - Medical Retrieval
retrieve.py - Answer Synthesis
run.py
Graph RAG Engine
- GraphRAG Runtime
nano_graphrag/graphrag.py - Graph Operations
nano_graphrag/_op.py - Graph Storage
nano_graphrag/_storage.py - LLM Adapter
nano_graphrag/_llm.py - Engine Base
nano_graphrag/base.py
CAMEL Framework
- Chat Agent
camel/agents/chat_agent.py - Task Model
camel/tasks/task.py - Workforce Manager
camel/workforce/manager_node.py - Workforce Worker
camel/workforce/worker_node.py - Task Channel
camel/workforce/task_channel.py - Vector Retriever
camel/retrievers/vector_retriever.py - Embeddings
camel/embeddings/base.py - Document Loader
camel/loaders/unstructured_io.py
Data And Storage
- Neo4j Graph Store
camel/storages/graph_storages/neo4j_graph.py - JSON Key Store
camel/storages/key_value_storages/json.py
Other components
- Questioner
- LLM Service
- PubMed Search
Connections 24
- Questioner → Run Workflow: starts
- Run Workflow → Medical Data Load: reads data
- Run Workflow → Document Chunking: chunks data
- Run Workflow → Graph Construction: builds graph
- Run Workflow → Graph Cleanup: merges graph
- Run Workflow → Graph Linking: links levels
- Run Workflow → Medical Retrieval: retrieves evidence
- Medical Retrieval → Answer Synthesis: supplies context
- Run Workflow → Answer Synthesis: runs inference
- Run Workflow → GraphRAG Runtime: uses runtime
- GraphRAG Runtime → Graph Operations: runs operations
nano_graphrag/graphrag.py - GraphRAG Runtime → Graph Storage: uses storage
nano_graphrag/graphrag.py - GraphRAG Runtime → LLM Adapter: calls LLM
nano_graphrag/graphrag.py - GraphRAG Runtime → Engine Base: uses base
nano_graphrag/graphrag.py - Graph Storage → JSON Key Store: uses key store
nano_graphrag/_storage.py - LLM Adapter → LLM Service: requests completion
- Run Workflow → PubMed Search: searches papers
- Vector Retriever → Embeddings: embeds query
camel/retrievers/vector_retriever.py - Vector Retriever → Document Loader: processes documents
camel/retrievers/vector_retriever.py - Workforce Manager → Chat Agent: creates coordinator
camel/workforce/manager_node.py - Workforce Manager → Task Channel: posts tasks
camel/workforce/manager_node.py - Workforce Worker → Task Channel: gets assigned tasks
camel/workforce/worker_node.py - Workforce Worker → Task Model: processes tasks
camel/workforce/worker_node.py - Medical Retrieval → Neo4j Graph Store: queries graph