How tanisharaj01/intellidocs-ai-powered-multi-document-intelligence-system works
IntelliDocs lets users upload PDFs, then ask questions and request summaries or cross-document comparisons with cited results. The primary path is Streamlit upload → FastAPI ingestion and embedding → chunk retrieval → answer generation → citation display; chat history is sent with questions. The README also documents hybrid vector/keyword search, reranking, abstention, and follow-up chat. Source excerpts show an in-memory backend and Gemini-based service, while OpenSearch support appears in retrieval code. Unsampled subsystem links are omitted where their runtime wiring is not evidenced.
Main entry points: main.py FastAPI endpoints · retrieval.py Hybrid retrieval and reranking · service.py RAG orchestration · app.py Streamlit workspace
Components 13
User interface
- Streamlit workspace
frontend/app.py
API and document lifecycle
- FastAPI endpoints
app/main.py - PDF parsing
app/rag/ingestion.py
RAG pipeline
- Local embeddings
app/rag/embeddings.py - RAG orchestration
app/rag/service.py - Hybrid retrieval and reranking
app/rag/retrieval.py - Answer generation
app/rag/generation.py - Authorization filters
app/rag/security.py
Index and models
- RAG data models
app/rag/models.py - Chunk index
app/rag/retrieval.py
Other components
- Document user
- Google Gemini
- OpenSearch service
Connections 15
- Document user → Streamlit workspace: uploads and asks
frontend/app.py - Streamlit workspace → FastAPI endpoints: sends API requests
frontend/app.py - FastAPI endpoints → PDF parsing: parses PDF
app/main.py - PDF parsing → RAG data models: creates chunks
app/rag/ingestion.py - FastAPI endpoints → Local embeddings: embeds uploaded text
app/main.py - FastAPI endpoints → Chunk index: adds chunks
app/main.py - FastAPI endpoints → RAG orchestration: delegates questions and summaries
app/main.py - RAG orchestration → Hybrid retrieval and reranking: retrieves candidates
app/rag/service.py - Hybrid retrieval and reranking → Local embeddings: scores vectors
app/rag/retrieval.py - Hybrid retrieval and reranking → Authorization filters: filters access
app/rag/retrieval.py - Hybrid retrieval and reranking → Chunk index: searches chunks
app/rag/retrieval.py - RAG orchestration → Google Gemini: generates answers
app/rag/service.py - RAG orchestration → RAG data models: uses RAG data types
app/rag/service.py - Streamlit workspace → Document user: shows answers and citations
frontend/app.py - Hybrid retrieval and reranking → OpenSearch service: supports backend
app/rag/retrieval.py