ModelMaestro

Inspiration

A lot of Data Scientists are tired of doing repetitive tasks of an ETL pipeline. Our project aims to automate that process so it eliminates the draining task, making it easier for them.

What it does

Our project takes in a raw data file and a prompt from the user explaining what the user's end goal is, after which our project orchestrates an entire Machine Learning pipeline by analyzing the data, selecting the best model to train, training the data, and producing a trained model along with a report explaining the model metrics.

How we built it

We built the backend using Google's Gemini to analyze the user's prompt and selecting the model while Vertex AI writes the code to orchestrate the pipeline. We created the frontend using three.js.

Challenges we ran into

The challenges we ran into were to fine tune the parameters and training time so the model is appropriately trained.

Accomplishments that we're proud of

To achieve the automated agentic workflow of the end-to-end Machine Learning Pipeline.

What we learned

We learned to gain an instinct of what types of models are usually best for what type of tasks and also see what data analyses looks right from a glance.