ModelPilot is an AI-powered platform that automates key stages of machine-learning experimentation. Users provide a dataset, select an objective, and define performance constraints. The system analyzes the dataset, runs controlled ML experiments, evaluates their results, and selects the best-performing candidate according to the specified objective. Our inspiration was the repetitive work involved in building reliable ML models: selecting metrics, testing configurations, comparing results, and deciding what to try next. We wanted to explore how an AI agent could help coordinate this process rather than simply generate code or answer questions. Using NVIDIA Nemotron through Nebius Token Factory for the reasoning layer, alongside a Python ML engine and a React frontend, ModelPilot aims to make experimentation more systematic, transparent, and accessible. Building the project taught us about ML evaluation, agent-tool architecture, API integration, deployment, experiment tracking, and the importance of designing AI systems around measurable outcomes.
Built With
- fastapi
- nemotron
- nvidia
- pandas
- python
- scikit-learn
- typescript
- vite