Data Science ML Agent

Overview

Data Science ML Agent is an AI-powered platform that combines a FastAPI backend with a Streamlit frontend to let users perform common data science workflows using plain English commands.

The project supports:

• Loading CSV and Parquet datasets
• Uploading dataset files via API
• Setting target columns and describing data
• Training classification and regression models
• Optimizing models with Optuna
• Saving and downloading the best model and predictions
• Tracking experiment history

Why it is useful

This repository is useful for developers and data scientists who want a conversational interface for fast prototyping and model exploration. It makes it easy to interact with datasets, train models, and inspect results without writing boilerplate code.

Getting started

Prerequisites

• Python 3.11+ (or compatible Python 3.x)
• pip package manager
• NVIDIA-compatible API key in environment variable NVIDIA_API_KEY

Install dependencies

python3 -m pip install --upgrade pip
python3 -m pip install -r requirements.txt

Configure the environment

Create a .env file in the repository root or export NVIDIA_API_KEY in your shell:

export NVIDIA_API_KEY="your_api_key_here"

The application also reads app/config.yaml for defaults such as base_url, app_name, and model_name.

Run the backend

Start the FastAPI server from the repository root:

fastapi dev app/main.py

This exposes the API on http://127.0.0.1:8000 and provides interactive docs at http://127.0.0.1:8000/docs.

Run the frontend

In a separate terminal, launch the Streamlit UI:

streamlit run frontend/app.py

The UI connects to the backend and allows chat-style commands, file upload, and downloads.

Usage examples

Use natural language commands in the Streamlit chat interface or send them to the /chat API endpoint.

Example commands:

load_dataset("data/train.csv", "target")
set_target("target")
describe_data()
preview_data(5)
train_classification()
train_regression()
optimize_logistic(20)
optimize_forest_regressor(30)
show_best_model("accuracy")
show_history(5)
predict("data/test.csv", "predictions.csv")
help()

API endpoints

The backend exposes the following operations:

• POST /chat — send a chat command and receive an AI-paraphrased response
• POST /upload — upload dataset files
• GET /history — get recent experiment history
• POST /clear — clear conversation history
• GET /download/model — download the best trained model
• GET /download/predictions — download generated prediction results

Project structure

• app/main.py — FastAPI backend entrypoint
• app/agents/chat_agent.py — conversational ML agent and tool registry
• app/services/llm.py — LLM client configuration and API integration
• app/agents/tools/ — data loading, model training, optimization utilities
• frontend/app.py — Streamlit user interface
• requirements.txt — Python dependencies
• data/ — example or sample dataset files

Help and support

• Use the built-in API docs at http://127.0.0.1:8000/docs
• Open an issue in the repository for bugs or feature requests
• Review LICENSE for licensing details

Contributing

Contributions are welcome via issues and pull requests. Keep changes focused on features, bug fixes, and documentation improvements.

License

This project is released under the MIT License. See LICENSE for details.