fact-checker

RAG system that lets you ask questions about Tim Walz and receive concise fact-checking answers with sources. It leverages a FAISS index to retrieve relevant context from pre-collected statements and a locally saved language model (using Hugging Face Transformers) to generate responses. The user interface is built with Streamlit.

Features

• FAISS Retrieval: Efficiently retrieves the most relevant context statements using vector similarity search.
• LLM: Uses a language model to generate fact-checking answers.
• Streamlit Chat Interface: Provides an interactive, chat-style UI for asking questions and viewing responses.
• Concise Output: Processes the generated answer to display only the final YES/NO decision along with one key supporting statement.

Installation

1. Clone the repository

git clone https://github.com/yourusername/fact-checker.git

2. Install dependencies

pip install -r requirements.txt

3. Run build_index.py script for creating and saving an efficient similarity search index using FAISS.

python build_index.py

4. Downloads a model and tokenizer from Hugging Face, then stores them in a local directory

python save_model.py

Running Locally

Launces app.py and fact_checker.py

streamlit run /Users/hardikgupta/Documents/Projects/model/app.py

Troubleshooting

Confirm that the directory contains the necessary files (faiss_index.bin, statements.pkl, my_local_model/). If not, follow the instructions again provided in the project documentation to generate or download these files.

Contributions are welcome!