A small, open source CNN. Built in plain Javascript, trained in Python. Runs in the browser, around 1700 parameters, not very accurate. It was a learning experience on how CNN's work, and I made the site for others to learn how they work.
WIP
A website and iOS app.
A javelin throw clip goes in. Each frame gets 2D keypoints, then a monocular 3D pose, then temporal smoothing. Out comes pose JSON, a overlay on the video, and a short report. Using custom made models for specific features (e.g. javein and back arm), and pretrained models for basic pose detection.
Working on: javelin detection, throw-phase detection, comparing two throws, and AI powered coaching.
WIP
A browser extension for Sparx Maths. Reads the question, fills in the answer, and remembers bookwork codes. Fully automates Sparx Maths.
An iOS game. Everyone gets a word, one player has a different one. People try to guess which player has the different word. Play it in the browser.