HackRx App

A FastAPI-based application for the HackRx challenge, designed to process documents and answer questions using embeddings and QA models.

Overview

This app uses:

• all-MiniLM-L6-v2 for embeddings.
• distilbert-base-uncased-distilled-squad for question answering.
• Pinecone for vector storage.
• PyPDF2 for document parsing.

Setup

1. Install Dependencies:

   .\hackrx_env\Scripts\Activate.ps1
   pip install -r requirements.txt

2. Pre-download Models (run once):

   python -c "from transformers import AutoTokenizer, AutoModel, pipeline; AutoTokenizer.from_pretrained('distilbert-base-uncased'); AutoModel.from_pretrained('sentence-transformers/all-MiniLM-L6-v2'); pipeline('question-answering', model='distilbert-base-uncased-distilled-squad')"

3. Run Locally:

   uvicorn main:app --host 0.0.0.0 --port 10000

Deploy with ngrok

1. Download ngrok from ngrok.com.
2. Start ngrok:

   .\ngrok http 10000

Notes

• Tested locally with ~2-4s response times (4.23s, 2.34s, 708ms).