Factify: AI-Powered Scientific Fact Checker

Project Overview

Factify is an advanced real-time fact-checking platform designed to verify scientific claims by leveraging cutting-edge artificial intelligence and scholarly databases. Our application provides researchers, students, journalists, and curious minds with evidence-based verification of scientific statements through a modern, responsive interface.

Key Features

• Instant Claim Verification: Enter any scientific claim and get real-time analysis of its validity backed by academic research
• Research-Backed Results: Access evidence from peer-reviewed studies across multiple academic databases
• AI-Powered Analysis: Utilizes advanced AI models to analyze claims and extract relevant information
• Citation Support: View and explore supporting research papers with full citation details and relevance scoring
• User-Friendly Interface: Clean, intuitive design with responsive and interactive UI elements

Technology Stack

Frontend

• React 19 with Vite for optimized builds and HMR
• React Router for client-side navigation
• Custom CSS with responsive design principles
• Interactive Magnetic UI elements for enhanced user experience
• Jest and Vitest for component testing

Backend

• Flask (Python) RESTful API with CORS support
• SQLAlchemy ORM with comprehensive database models
• Supabase for authentication, storage, and PostgreSQL database
• FAISS vector database for high-performance similarity search
• Sentence Transformers for semantic text embeddings
• Concurrent processing with thread pooling for faster responses

AI/ML Components

• RAG Architecture: Implements a sophisticated Retrieval Augmented Generation pipeline for accurate fact verification
• Semantic Embeddings: Utilizes the all-MiniLM-L6-v2 model (384-dimensional embeddings) for document and query representation
• Vector Search: FAISS IndexFlatL2 for efficient L2 distance-based nearest neighbor search
• LLM Integration: Google's Generative AI (Gemini) with prompt engineering for:
  • Claim decomposition into atomic verifiable statements
  • Evidence synthesis across multiple research papers
  • Confidence scoring with numerical justification
  • Citation and reference formatting
• Adaptive Knowledge Retrieval: Dynamically adjusts retrieval strategy based on claim complexity
• Memory Optimization: Implements lazy loading for ML models and memory-efficient batching

External APIs

• OpenAlex API for comprehensive academic paper retrieval with citation metrics
• CrossRef API for cross-referencing academic sources
• Semantic Scholar API for additional paper information and citation contexts

Getting Started

Prerequisites

• Node.js (v18+)
• Python (v3.10+)
• pip (Python package manager)
• npm (Node package manager)
• Supabase account (for database and auth)

Installation

1. Clone the repository:

   git clone https://github.com/CMPT-276-SPRING-2025/final-project-15-mountains.git
   cd final-project-15-mountains

2. Set up the backend:

   cd src/backend

   For Linux/macOS:

   pip install -r requirements.txt

   For Windows users: faiss-cpu cannot be installed via pip on Windows by default. Please follow these steps:

   • First, install faiss-cpu via conda:

     conda install -c pytorch faiss-cpu

   • Then install the remaining dependencies:

     pip install -r requirements.txt --no-deps

   • Copy and update your environment variables:

     cp .env.example .env
     # Update the .env file with your API keys

3. Set up the frontend:

   cd src/frontend
   npm install

4. Run the application:

   # In one terminal (from the src directory)
   python backend/app.py
   
   # In another terminal (from the src directory)
   npm run dev

5. Open your browser and navigate to http://localhost:5173

License

This project is licensed under the MIT License - see the LICENSE file for details.

Copyright (c) 2025 Factify Team