Product & Engineering
Building at the edge of research and product — machine learning, interfaces, and the systems that hold them together.
I work where research meets shipped product — training models one week, shaping the interface that makes them useful the next. Past work spans generative reconstruction of lunar terrain, quantized networks for compressed image recovery, and consumer mobile products. I care about systems that stay legible as they get complex.
A nutrition companion that turns a photo of a meal into structured macro data through an on-device vision pipeline and a nutrition inference service. Built end to end — data model, scanning flow, streaks, and the goal engine behind the weekly progress view.
Generative adversarial models trained to interpolate between sparse orbital captures, synthesising continuous high-resolution lunar terrain. The work explores how latent-space traversal preserves geological structure — crater rims, ridges, and shadow geometry — where classical interpolation smears it away.
Compressed image recovery with unrolled proximal averaging networks whose weights and activations are quantized, cutting memory and compute while holding reconstruction fidelity. Reconstructions stay sharp at 2 and 3 resolvent blocks, with lower residual error than quantized ISTA-Net.
