Alex Dils Computer Vision · Medical AI
Vol. 03 / No. 02
Berkeley · 2026
§ Story

How I started caring less about scores, and more about misses.

A short note on how I got into computer vision, medical AI, and the simulation-driven side of augmentation.

01The hook

Vision exposes failure quickly.

A model can look good on clean data and still break on small shifts.

I got into vision because it exposes failure quickly. I started caring less about benchmark scores and more about what changes in the world cause a miss.

02Stanford, 2022

Add structure to training, not noise.

In 2022 I joined Stanford Biomedical Informatics as a research intern. I worked on confounders in skin lesion diagnosis and on physiologically plausible augmentation for tumor segmentation.

The core idea was to add structure to training, not noise.

03Shipping

Shipping forces discipline.

On the systems side, I helped build Appraise AI. Shipping a model forces discipline. It makes you name failure modes, track drift, and communicate uncertainty without hiding behind averages.

← Back to index.