How maheshm-01/multimodel_graph_rag works
This is a multimodal Graph RAG platform: users upload documents or ask grounded questions through a web dashboard or API. Ingestion loads and chunks documents, extracts graph entities, embeds content, and indexes it; query handling routes intent, retrieves across hybrid channels, reranks evidence, and synthesizes cited answers. The graph preserves the documented visualizer and job-status capabilities. Some subsystem links rely on README or SOURCE INDEX references because their implementations were not excerpted; those limits are reflected by omitted or dashed edges.
Main entry points: ingestion_pipeline.py Ingestion Workflow · query.py Chat Query API · app.py FastAPI App & Routes · hybrid_fusion.py Hybrid Retrieval
Components 26
User Experience & API
- Web Dashboard
src/static/js/app.js - FastAPI App & Routes
src/api/app.py - Ingest & Jobs API
src/api/v1/endpoints/ingest.py - Chat Query API
src/api/v1/endpoints/query.py - Graph API
src/api/v1/endpoints/graph.py
Workflow Orchestration
- Background Worker
src/worker.py - Ingestion Workflow
src/workflows/ingestion_pipeline.py - RAG Workflow
src/workflows/rag_pipeline.py
Ingestion & Retrieval
- Document Loaders
src/ingestion/loaders - Multimodal Chunking
src/ingestion/chunkers - Graph Extraction
src/graph/extraction/extractor.py - Graph Construction
src/graph/construction/builder.py - Intent Router
src/retrieval/query_router.py - Hybrid Retrieval
src/retrieval/hybrid_fusion.py - Document Search
src/retrieval/document_search.py - Multimodal Reranker
src/retrieval/reranker.py - Grounded Synthesizer
src/generation/synthesizer.py
Stores & Models
- Text & Vision Embeddings
src/embeddings - Vector Store
src/storage/vector_store.py - Graph Store
src/storage/graph_store.py - Domain Models
src/domain - Application Settings
src/config/settings.py
Other components
- User
- Neo4j
- Qdrant
- LLM / VLM Providers
Connections 33
- User → Web Dashboard: uses
README.md - Web Dashboard → FastAPI App & Routes: requests
- User → FastAPI App & Routes: calls
README.md - FastAPI App & Routes → Ingest & Jobs API: routes to
- FastAPI App & Routes → Chat Query API: routes to
- FastAPI App & Routes → Graph API: routes to
- Ingest & Jobs API → Background Worker: registers jobs
src/api/v1/endpoints/ingest.py - Background Worker → Ingestion Workflow: executes
src/worker.py - Ingestion Workflow → Document Loaders: loads documents
src/workflows/ingestion_pipeline.py - Ingestion Workflow → Multimodal Chunking: chunks content
src/workflows/ingestion_pipeline.py - Ingestion Workflow → Graph Extraction: extracts entities
src/workflows/ingestion_pipeline.py - Ingestion Workflow → Graph Construction: builds graph
src/workflows/ingestion_pipeline.py - Ingestion Workflow → Text & Vision Embeddings: embeds chunks
src/workflows/ingestion_pipeline.py - Ingestion Workflow → Vector Store: indexes vectors
src/workflows/ingestion_pipeline.py - Graph Construction → Neo4j: writes graph
- Chat Query API → Intent Router: routes intent
src/api/v1/endpoints/query.py - Chat Query API → Hybrid Retrieval: retrieves evidence
src/api/v1/endpoints/query.py - Chat Query API → Document Search: searches documents
src/api/v1/endpoints/query.py - Hybrid Retrieval → Graph Store: searches graph
src/retrieval/hybrid_fusion.py - Hybrid Retrieval → Vector Store: searches vectors
- Hybrid Retrieval → Text & Vision Embeddings: uses embeddings
- Chat Query API → Multimodal Reranker: reranks results
src/api/v1/endpoints/query.py - Chat Query API → Grounded Synthesizer: synthesizes answer
src/api/v1/endpoints/query.py - Grounded Synthesizer → LLM / VLM Providers: generates response
src/generation/synthesizer.py - Graph Store → Neo4j: queries graph
src/storage/graph_store.py - Vector Store → Qdrant: stores vectors
- RAG Workflow → Intent Router: routes queries
src/workflows/rag_pipeline.py - RAG Workflow → Hybrid Retrieval: retrieves
src/workflows/rag_pipeline.py - RAG Workflow → Multimodal Reranker: reranks
src/workflows/rag_pipeline.py - RAG Workflow → Grounded Synthesizer: synthesizes
src/workflows/rag_pipeline.py - FastAPI App & Routes → Application Settings: configures
src/api/app.py - Ingestion Workflow → Domain Models: uses job and document models
src/workflows/ingestion_pipeline.py - Chat Query API → Domain Models: uses request models
src/api/v1/endpoints/query.py