Skip to content

A chatbot for Kaggle competition inquiries, using LangChain for processing, a Mistral model for responses, and Retrieval-Augmented Generation (RAG) for dynamic data. It features a FastAPI backend and a Streamlit front end for real-time interaction.

Notifications You must be signed in to change notification settings

kaoutaar/KaggleBOT

Repository files navigation

Kagil BOT

A bot that answers your questions about kaggle competitions, choose a competittion in the listBOX, let the backend program gather the necessary data from kaggle website, and here you go! the bot is ready to receive your questions.


Screenshot (84)

Screenshot (87)


How to run:

  1. clone the repo
  2. download the kaggle json file containing your credentials and copy it to your local directory ~/.kaggle
  • on Windows:

    pip install the requirements and open 2 separate cmd terminals, cd to the repo directory and run:

    • cmd1: python back_app.py
    • cmd2: streamlit run app.py
  • on WSL:

    in a shell terminal, cd to the repo directory and run: bash start.sh

Under the hood:

arch

after picking a competition, all related data including discussions and notebooks sections get loaded then splitted and stored in faiss vectorstore. we use mistral-7b-gguf model, which can totally run on cpu at the cost of being slow, the model is served using a light weight API.

limitation:

with 8gb RAM, the model takes about 120 seconds to generate a full answer.

About

A chatbot for Kaggle competition inquiries, using LangChain for processing, a Mistral model for responses, and Retrieval-Augmented Generation (RAG) for dynamic data. It features a FastAPI backend and a Streamlit front end for real-time interaction.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published