Hackathon
Aphasia Annotate
A multimodal agent that helps speech-language therapists annotate gesture and aphasic speech from video, and asks when it isn't sure instead of guessing.
Industry research and things I've built myself, each with the full process: how I framed the problem, what I tried, what I found and what I'd do next. For my PhD and current academic work, see Research.
Hackathon
A multimodal agent that helps speech-language therapists annotate gesture and aphasic speech from video, and asks when it isn't sure instead of guessing.
Referencewhat was said
terusakunggakisomulihsoaleudanderes
Baseline79.2% WER
terusakunggakisaМулихsoalnyaudanderesya
After diagnosis35.4% WER
terusakunggakisomulihsoaleudahderes
Hackathon
A speech recognition model kept transcribing Javanese conversations in Cyrillic and Chinese script. Reading its mistakes, rather than reaching for a bigger model, cut the error rate from 79% to 35%.
Industry
A remote field study of an audio-visual podcast player, testing whether images, transcripts and chapters help or get in the way in everyday listening.
Personal
An AI tool that turns weather forecasts into plain-language planting advice for smallholder corn farmers in Java, Indonesia.
Personal
A map of where and when cherry blossoms are blooming across London, built quickly with AI and checked against what people were posting in real time.