Active prototype
Longitudinal Health Data Platform
A non-diagnostic prototype for following clinical measurements, personal observations and wearable data over time. The aim is to make long-term patterns easier to inspect without presenting the system as a medical authority.
I use the project to explore data quality, trends, uncertainty and risk communication. It is intended to help organise information and raise useful questions for discussion with qualified clinicians, not to diagnose disease or replace healthcare.
Active prototype
Skin Change Monitoring
This project began as a method for creating reproducible time-separated images of the skin. The aim is to reduce variation caused by camera position, scale and lighting, then use computer vision to flag visible changes for manual review.
A separate machine-learning track tests classification methods trained on open research datasets. The output is treated as an experimental signal, not a diagnosis. I am particularly interested in dataset limitations, false reassurance, false alarms and whether results can be reproduced.
Early construction phase
Companion Robot and Mobile Sensor Platform
I am rebuilding a domestic robot platform as a small mobile testbed for robotics, smart-home integration and interaction with people.
The daytime idea is a playful companion that can take part in simple games, use light and sound expressively, and support supervised voice interaction. It may also collect environmental readings and run a small number of predefined household routines.
Movement and safety stay under local, rule-based control. AI services may interpret speech, examine selected images or choose from a limited set of allowed actions, but they do not get unrestricted access to the motors. Camera and microphone activity should be obvious to people nearby and easy to disable.
Supporting infrastructure
Distributed AI Development Lab
I use a small collection of computers with different capabilities for software development, model training and testing on less powerful target hardware.
Code, tests and model files are versioned as they move between machines. Models are trained where the necessary hardware is available, then tested again on the device that is expected to run them. This gives me a practical way to learn deployment, test automation and reproducibility.
GitPythonTDDModel deploymentEdge validationMLOps learning