An agent that helps analyze and output insights based on news sources, using TogetherAI and Streamlit. This project is meant to help extract and summarize viewpoints about the 2020 Presidential Election from articles in both English and Spanish.
This project helps implement an LLM-based QA system that helps users with questions about the election. Specifically, this system is able to help with cross-lingual analysis.
- grounds the system's answers using article text
- identifies the two most relevant articles for each language
- uses FAISS indexing to efficiently identify relevant articles
- determines if a query is relevant before the system retrieves articles
- when applicable, uses the past two interactions in chat history to determine this
- easy to use Streamlit UI
- chat history and quick, concise responses
- LLM based evaluation
- creates logs for each day of usage
-
Obtain a dataset of your choosing of news articles. We retrieved articles based on work done here: https://huggingface.co/datasets/stanford-oval/ccnews.
-
Clone the repository
-
Install dependencies using a virtual environment
pip install -r requirements.txt
-
Obtain a Together API key in a config.json file
-
Adjust filtering and prompt based on your use case
-
Run the Streamlit app:
streamlit run run.py