Machine learning researcher & engineer
Computer vision · generative models · speech · and the evaluation of all three
I build and evaluate models across computer vision, speech, generative modelling and reinforcement learning, with applied work in clinical research and quantitative finance.
I like problems with something real pushing back a model that has to run on a small device, or one that has to hold up across a corpus far too large to eyeball. Scarcity and scale are interesting for the same reason: neither lets you hand-wave.
I show people the rough version. Quick ugly baseline, shared before it's presentable. A half-built thing someone can react to beats a polished thing built on an assumption nobody checked.
I ask the obvious question. Cheerfully, and twice if needed. Thirty seconds of looking slow costs less than a fortnight on the wrong problem — and usually someone else in the room was wondering too.
I work backwards from whoever has to use it. A number isn't an outcome. Somebody acts on it, under time pressure, without reading the methods section. That belongs in the conversation on day one, not at the write-up.
I love to discuss. Most of the useful corrections I've had came from clinicians, statisticians and quants who didn't share my priors. Change my mind and I'll say so out loud.
If that sounds like a fit — say hello. I'm always up for a conversation about a problem I don't know anything about yet.
🔬 Some of the things I've worked on
Medical imaging — offline sonography models for breast screening on hand-held hardware; oral-cancer classification; unsupervised cancer detection; X-ray enhancement with DCGAN and pix2pix.
Foundation models & evaluation — real-time monitors built on frozen backbones; generative model evaluation on Riemannian manifolds; speech-model robustness under corrupted audio.
Sequence & signal — human activity recognition under domain shift; WiFi-CSI liquid neural networks; equity forecasting with Bi-LSTM and CNN-LSTM.