I build machine learning that has to work outside a notebook.
A model that scores well in a notebook has not finished the job. Mine run on a Pi, behind an API, against data nobody cleaned first.
Pose estimation and movement analysis, on video and on constrained hardware.
Getting useful answers out of records that were never meant to be queried.
The pipelines, services and deployment that keep both of those standing.
Scores a gymnastics routine from video, phase by phase.
Answers questions across production data held in spreadsheets.
Reads sign language letters from a live webcam, across two machines.
An agent learns FEI-inspired courses in Unity ML-Agents.
Footage in, optimised signal timings out.
Movement analysis, posture and lameness detection. Most of my ideas start at a stable, which is how a reinforcement learning horse ended up in my portfolio.
Further out: quantum computing, and the mathematics and physics underneath all of this. That is the research I would like to spend my career on.