Inspiration
We wanted an easier way to visualize and learn about Neural Networks. Taking from virtual code playgrounds like Scratch, we built Stitch to provide users with an interactive learning experience.
What it does
Users can build different models, learn what each component does, adjust hyperparameters and observe these changes in training. Users can then test their models.
How we built it
This project combines a React/TypeScript front end with a FastAPI/PyTorch back end to let users build, train, inspect, and test MNIST digit classifiers. It powers an interactive UI where you can draw digits, launch hyper‑parameter sweeps, inspect layer architectures, and visualize how pixels flow through the network.
- Frontend: React 19 + Vite + Tailwind + React Flow for rich network visualizations.
- Backend: FastAPI + PyTorch for model management, training, and inference.
- Storage: Lightweight file system persistence (pickled PyTorch models plus JSON metadata).
- Tooling: pnpm for front-end dependencies, uv/pip for Python, and React Query for data fetching.
Built With
- flask
- python
- pytorch
- react
- tailwindcss
- typescript
Log in or sign up for Devpost to join the conversation.