ankitkumarai/ragfusion-multimodel-private-rag--backend

How ankitkumarai/ragfusion-multimodel-private-rag--backend works

This is a private, owner-scoped multimodal RAG application exposed through FastAPI and a CLI. Its main workflow accepts an authenticated user’s upload, stores the original and metadata, extracts and chunks supported content, embeds it into Qdrant, then retrieves authorized chunks for grounded answer generation. The README also documents authentication, conversation history, document lifecycle and recovery, profile photos, evaluation, and local or Supabase storage. Edges are limited to README statements and shown source references; unsampled subsystem wiring is left unasserted where it cannot be verified.

Main entry points: documents/service.py Document lifecycle · ingestion.py Ingestion orchestration · app.py FastAPI · rag/service.py RAG service

Components 25

API and identity

FastAPI src/multimodal_private_rag/api/app.py
Authentication src/multimodal_private_rag/auth/service.py
User identity src/multimodal_private_rag/auth/models.py
Chat routes src/multimodal_private_rag/api/routers/chat.py
Conversation history src/multimodal_private_rag/conversations/service.py
Document routes src/multimodal_private_rag/api/routers/documents.py
Profile photos src/multimodal_private_rag/auth/photos.py

Document lifecycle

Document lifecycle src/multimodal_private_rag/documents/service.py
Ingestion orchestration src/multimodal_private_rag/rag/ingestion.py
Format extraction src/multimodal_private_rag/ingestion
Chunking src/multimodal_private_rag/ingestion/chunking.py

Retrieval and answers

RAG service src/multimodal_private_rag/rag/service.py
Candidate retrieval src/multimodal_private_rag/retrieval/retriever.py
Cross-encoder reranking src/multimodal_private_rag/reranking/fastembed.py
Context selection src/multimodal_private_rag/rag/context.py
Answer generation src/multimodal_private_rag/generation/gemini.py

Persistence and providers

Embeddings src/multimodal_private_rag/embeddings/fastembed.py
Qdrant vectors src/multimodal_private_rag/vectorstore/qdrant.py
Document metadata src/multimodal_private_rag/documents/sqlite.py
Original-file storage src/multimodal_private_rag/storage/base.py
Supabase Storage src/multimodal_private_rag/storage/supabase.py
PostgreSQL src/multimodal_private_rag/persistence/database.py

Operations and evaluation

RAG evaluation src/multimodal_private_rag/evaluation
Lifecycle recovery src/multimodal_private_rag/operations.py

Other components

User

Connections 28

User → FastAPI: uses API README.md
FastAPI → Chat routes: registers routes src/multimodal_private_rag/api/app.py
FastAPI → Document routes: registers routes src/multimodal_private_rag/api/app.py
FastAPI → Authentication: registers routes src/multimodal_private_rag/api/app.py
Authentication → User identity: establishes identity src/multimodal_private_rag/api/routers/auth.py
Chat routes → RAG service: requests answers
Chat routes → Conversation history: uses history
Document routes → Document lifecycle: handles documents
Document lifecycle → Original-file storage: stores originals README.md
Document lifecycle → Document metadata: persists metadata
Document lifecycle → Qdrant vectors: deletes vectors src/multimodal_private_rag/documents/service.py
Document lifecycle → Ingestion orchestration: processes uploads src/multimodal_private_rag/documents/service.py
Ingestion orchestration → Format extraction: loads content src/multimodal_private_rag/rag/ingestion.py
Ingestion orchestration → Chunking: chunks content src/multimodal_private_rag/rag/ingestion.py
Ingestion orchestration → Embeddings: embeds chunks src/multimodal_private_rag/rag/ingestion.py
Ingestion orchestration → Qdrant vectors: upserts vectors src/multimodal_private_rag/rag/ingestion.py
RAG service → Candidate retrieval: retrieves candidates src/multimodal_private_rag/rag/service.py
Candidate retrieval → Embeddings: embeds query src/multimodal_private_rag/retrieval/retriever.py
Candidate retrieval → Qdrant vectors: searches vectors src/multimodal_private_rag/retrieval/retriever.py
Candidate retrieval → Cross-encoder reranking: reranks candidates src/multimodal_private_rag/retrieval/retriever.py
RAG service → Context selection: selects context src/multimodal_private_rag/rag/service.py
RAG service → Answer generation: generates answer src/multimodal_private_rag/rag/service.py
Original-file storage → Supabase Storage: optional adapter README.md
Document lifecycle → Profile photos: separate from RAG README.md
User → RAG evaluation: runs evaluation README.md
Lifecycle recovery → Document metadata: reconciles lifecycle README.md
Lifecycle recovery → Qdrant vectors: reconciles integrity README.md
PostgreSQL → Document metadata: stores production metadata README.md

Drawn by GitDiagram on November 13, 2026 from the repository's file tree, README and source files. Read as Markdown ·