core Estimated learning time: 20 h

13.2 Capstone 2 — vision or audio system

A working perception system with a demo people can try.

Before:05. Classical Machine Learning

The second capstone ships a perception system — vision or audio — with transfer learning, latency optimisation and a public demo people can actually try, ideally including some data you collected yourself, which teaches lessons no downloaded dataset can. It sits second because it cashes in the deep learning module. Stopping at the notebook is the failure mode: an interactive demo on a public Space is what turns a project into evidence.

Work through these

  • Dataset collection or curation, including your own data

    Assembling a dataset, including collecting some of it yourself, which teaches more about data than any public set can. It is also what makes the project distinctively yours.

  • Transfer learning and augmentation

    Starting from a pretrained model and expanding the data artificially, which is what makes a strong result possible on a small collection. These two techniques carry the whole project.

  • Latency and size optimization for deployment

    Making the model small and fast enough to serve interactively. The constraint is what turns a notebook result into a system.

  • Interactive demo (Gradio or Streamlit) on HF Spaces

    An interactive demonstration hosted publicly, so the work can be tried rather than described. This is the single most effective portfolio item in this module.

Sign in to keep your progress.

Free resources

Links last checked 29 Aug 2026.

Stuck here?

Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.

Checking your session…

Topics shown in module order.