salvirezwan/Research-Paper-RAG-chatbot

Research Paper RAG Chatbot

An AI-powered agentic RAG (Retrieval-Augmented Generation) system for students and researchers. Upload research papers or fetch them live from arXiv, then query across them in natural language with cited, context-aware responses — streamed in real time.

Live Demo: Hugging Face Spaces

Features

• —Upload PDFs — ingest your own research papers with a 4-step checkpointed pipeline
• —Fetch from arXiv — download and index papers directly by arXiv ID
• —Agentic RAG pipeline — LangGraph StateGraph with adaptive routing, document grading, and cited answer generation
• —Real-time streaming — chat responses streamed via Server-Sent Events (SSE)
• —Session isolation — each browser session has its own paper library and vector search scope
• —PDF Viewer — read papers in-browser with page navigation

Architecture

RAG Pipeline (LangGraph StateGraph)

User Query → Router → [retrieve | live_fetch] → grade_docs → generator → citation → END

Ingestion Pipeline (4-step, checkpointed)

PDF → Parse (PyMuPDF) → Clean → Chunk → Embed (BAAI/bge-base-en-v1.5) → ChromaDB

Each step is checkpointed in MongoDB. Retrying a failed ingestion skips already-completed steps.

Storage

API Routes

Quick Start (Local)

Prerequisites

• —Python 3.11+
• —MongoDB running on localhost:27017
• —uv (recommended) or pip

1. Clone the repo

git clone https://github.com/salvirezwan/Research-Paper-RAG-chatbot.git
cd "Research-Paper-RAG-chatbot/Academic Research RAG"

2. Install dependencies

# Using uv (recommended)
pip install uv
uv sync

# Or using pip
pip install -r requirements.txt

3. Configure environment

cp .env.example .env

Edit .env:

GROQ_API_KEY=your_groq_api_key_here
MONGODB_URL=mongodb://localhost:27017
MONGODB_DATABASE_NAME=academic_research_rag
CHROMA_PERSIST_PATH=./data/chroma_db
CHROMA_COLLECTION_NAME=research_papers
GROQ_MODEL=llama-3.3-70b-versatile
EMBED_MODEL_NAME=BAAI/bge-base-en-v1.5
UPLOAD_DIR=uploads/documents

Get a free Groq API key at console.groq.com

4. Run

# Option A — dev shortcut (Windows, opens two terminals)
.\dev.bat

# Option B — manual
python -m uvicorn backend.main:app --host 0.0.0.0 --port 8000 --reload
streamlit run frontend/app.py

• —Backend: http://localhost:8000
• —Frontend: http://localhost:8501
• —API Docs: http://localhost:8000/docs

Docker (Full Stack)

docker compose -f docker/docker-compose.yml up --build

• —Frontend: http://localhost:8501
• —Backend: http://localhost:8000

Hugging Face Spaces Deployment

The app runs as a single Docker container on a free HF Space (2 vCPU / 16 GB RAM).

Process layout

supervisord (PID 1)
├── nginx        → port 7860  (reverse proxy)
│   ├── /api/*  → 127.0.0.1:8000  (FastAPI)
│   └── /*      → 127.0.0.1:8501  (Streamlit)
├── uvicorn      → port 8000
└── streamlit    → port 8501

Ephemeral storage

Since HF Spaces has no persistent disk, the app uses:

• —ChromaDB EphemeralClient (in-memory vectors)
• —mongomock-motor AsyncMongoMockClient (in-memory MongoDB)
• —/tmp/uploads/ for uploaded files

All data is lost on restart — expected behaviour for the free tier.

Deploy your own

1. 1.Create a new Space at huggingface.co → Docker SDK
2. 2.Push this repo to the Space's git remote
3. 3.Space Settings → Secrets: add GROQ_API_KEY
4. 4.Space Settings → Variables: add APP_PUBLIC_URL = https://<your-username>-<your-space-name>.hf.space
5. 5.First startup takes ~5 min (downloads the ~450 MB embedding model)

Environment Variables

Tech Stack