AutoML: Chat-Driven Machine Learning evaluations

This project reimagines AutoML through conversation. Built with GPT-3.5 and LangChain, the chatbot converts natural language prompts into JSON-formatted model training workflows using Scikit-learn. From hyperparameter tuning to real-time performance metrics, it turns casual dialogue into fully functioning ML models — no coding, just results. Ideal for rapid prototyping and ML education.

The Problem

Building and tuning machine learning models still requires a fair bit of technical know-how — not to mention scripting, debugging, and experimenting with different parameters. For non-experts, even small ML tasks can feel out of reach, and for developers, repetitive experimentation eats into valuable time.

The Solution

We designed a conversational chatbot that turns natural language prompts into working ML pipelines for model training and evalutions. Powered by GPT-3.5 and LangChain, our chatbot translates user queries into JSON workflows, trains models using scikit-learn, and returns performance metrics and plots — no coding required.

From hyperparameter tuning to confusion matrices, the chatbot handles everything end-to-end. Users simply describe the task (“Train a Random Forest on churn.csv with 30% test split”), and the system builds, trains, evaluates, and returns results visually and textually.

This made ML experimentation dramatically faster and more accessible — not only reducing friction for new learners, but helping developers iterate quicker by automating repetitive workflows. All evaluation outputs (e.g. F1 scores, confusion matrices) were displayed both in logs and charts to aid in interpretation.

Tech Stack

• Language Model: GPT-3.5 along with LangChain powered RAG system

• Backend ML Engine: Scikit-learn

• Workflow Generation: JSON-based model training pipelines

• Output Visualisation: Matplotlib, seaborn

• Development Practices: Agile collaboration, CI/CD

For more details, checkout my Github page: https://github.com/pparashar21/AutoML_GPT