PaperSense – PDF Question Answering System using RAG and Gemini

PaperSense is a Retrieval-Augmented Generation (RAG) system that allows users to upload research-relevant documents in PDF format and ask questions. The system retrieves relevant document sections using semantic search and generates grounded answers using Google's Gemini LLM, gemini-2.5-flash.

Features

• Upload research papers dynamically via REST API or UI • Automatic PDF parsing, chunking, and vector embedding • Semantic retrieval using FAISS • Context-aware question answering using Gemini • Source citation for every answer • Minimal and clean Streamlit user interface

System Architecture

1. PDFs are uploaded through API or UI
2. Text is extracted and split into overlapping chunks
3. Chunks are embedded using Sentence Transformers
4. Embeddings are stored in a FAISS vector index
5. User queries retrieve the most relevant chunks
6. Gemini generates answers grounded in retrieved context

Tech Stack

• Backend API: Flask • LLM: gemini-2.5-flash • Embeddings: Sentence Transformers • Vector Store: FAISS • PDF Parsing: pypdf • UI: Streamlit

How to Run Locally

1. Clone the repository
2. Create a virtual environment
3. Install dependencies from requirements.txt
4. Add your own Gemini API key in a .env file
5. Run the Flask API
6. Run the Streamlit UI

Example Use Cases

• Research paper analysis • Literature review assistance • Academic project demos • RAG-based GenAI experimentation

Security Note

The .env file containing API keys is intentionally excluded. Users must provide their own API keys to run the application.

Project Name

PaperSense