Efficient RAG retrieval system for article fragments from the Kaggle dataset available here. This system supports popular vector stores retrieval and includes a Question Answering functionality with Large Language Models (LLMs). By default, it utilizes FAISS for vector store retrieval and the Mixtral-8x7B LLM. More details in report.

To download the dataset, you can choose one of two options:

1. Download it manually from link and create folder named data in project root directory. Then store the medium.csv file in that folder.
2. Use the Kaggle API: Download your account token from this link and overwrite the existing kaggle.json file.

This step is not obligatory but necessary if you want to use Q/A system with Large Language Model support. To obtain your HuggingFaceHub API Token generate it and copy it from your HuggingFace account and paste it to .env file overwriting <YOUR_TOKEN> placeholder.

pip install pipx
pipx install poetry

sudo snap install ollama
ollama pull llama3

litellm --model ollama/llama3

poetry install
poetry shell
chmod +x kaggle.sh
./kaggle.sh
streamlit run app.py

sudo docker build -t ars-app:latest .
sudo docker container run -it -p 8501:8501 ars-app:latest

Got to LangSmith website, sign in, generate API key and paste it into .env.

To be able to evaluate the system's answers you need to generate an OpenAI API key from link and paste it into .env.

To run application with observability mode:

LLMOPS=true streamlit run app.py