README.md

metadata

title: DocChat Backend
emoji: 📄
colorFrom: green
colorTo: blue
sdk: docker
pinned: false

Querify

RAG-powered document chat. Upload a PDF, TXT, or CSV and ask questions against its content using a conversational interface.

Features

• Chunked document ingestion with configurable size and overlap
• Knowledge graph indexing (entities + relations) for GraphRAG retrieval
• Retrieval-augmented generation using Qwen, Llama, or Mistral via Hugging Face Inference API
• Auto-recommended chunking parameters based on document size and type
• Async graph-index build progress tracking

Project Structure

doc_ingestion/
├── backend/
│   ├── main.py          # FastAPI routes
│   ├── ingestion.py     # File handling and metadata
│   ├── parser.py        # Text extraction (PDF, TXT, CSV)
│   └── rag.py           # Chunking, graph indexing, retrieval, generation
├── frontend/
│   └── app.py           # Streamlit UI
├── Dockerfile
└── requirements.txt

Quick Start

cd doc_ingestion
pip install -r requirements.txt

Set environment variables:

export HF_TOKEN=your_hf_token           # LLM inference

Start the backend:

uvicorn backend.main:app --reload --port 8000

Start the frontend (separate terminal):

cd frontend
streamlit run app.py

Open http://localhost:8501, upload a document, and start chatting.

API Reference

Method	Endpoint	Description
POST	/upload	Upload and ingest a file
POST	/ingest_text	Ingest raw text directly
GET	/ingest_status	Poll graph-index progress
GET	/graph_store_status	Graph index health and stats
POST	/query	Submit a question

Interactive docs available at http://localhost:8000/docs.

Configuration

Parameter	Default	Description
chunk_size	180	Characters per chunk
chunk_overlap	40	Overlap between adjacent chunks
top_k	3	Chunks retrieved per query