CoolFace
Apppublic

Abdu2005/research-paper-assistant

sourceHugging Faceupdated 5mo agoView on Hugging Face
0likes
App README

📖 AI Research Paper Assistant

An agentic RAG system that lets you upload research papers and ask questions with cited answers.

Features

  • —Multi-document support — Upload multiple PDFs and ask cross-paper questions
  • —Cited answers — Every answer shows source chunks with page numbers
  • —ArXiv web search — Autonomous agent finds related papers on the web
  • —LLM Router — Intelligently routes queries to RAG or web search
  • —RAGAS Evaluation — Built-in faithfulness scoring (0.63 on Transformer paper)
  • —Filter by Paper — Ask questions about specific papers or all at once

Tech Stack

ComponentTechnology
LLMGroq LLaMA 3.1 8B
EmbeddingsHuggingFace all-MiniLM-L6-v2
Vector StoreFAISS
FrameworkLangChain + Streamlit
Web SearchArXiv API
EvaluationRAGAS

How to Use

  1. 1.Upload one or more research paper PDFs
  2. 2.Ask any question about the paper(s)
  3. 3.Enable Web Search Agent to find related papers on ArXiv
  4. 4.Use Filter by Paper to ask about specific papers
  5. 5.Click Run Eval to see RAGAS faithfulness score

Local Setup

bash
git clone https://huggingface.co/spaces/YOUR_USERNAME/research-assistant
cd research-assistant
pip install -r requirements.txt

Create .env file:

Run:

bash
streamlit run app.py

Example Questions

  • —"What is the main contribution of this paper?"
  • —"Compare the architecture of both papers"
  • —"What does BERT say about pretraining?"
  • —"Find related papers on transformer architecture"

Evaluation Results

Tested on "Attention is All You Need" (Transformer paper):

  • —RAGAS Faithfulness: 0.63

Built with LangChain + Groq + FAISS + Streamlit