My running watch collects more data than I know what to do with. I built Plover to point out what changed each week, like a jump in mileage or a higher heart rate at my usual pace, and show the data behind it. I’ve only tested it with my own running data. It does not predict injuries.
Tested three picker designs with roughly six engineers.
Reduced selection of frequently used models from five clicks and a scroll to two clicks.
Amazon Bedrock lets developers build apps with AI models from different providers, but choosing a model meant looking through a list of names without knowing what each supported. I redesigned the picker to show that information alongside search and recently used models. It shipped across six parts of Bedrock IDE.
Password required
03
Reusable AI chat components for AWS
Frontend Engineer (research + prototyping) · Research prototype
Built an interactive documentation prototype to research demand across AWS teams.
Engineers built the package; parts now run in the Bedrock playground.
Teams across Amazon were building similar AI chat interfaces from scratch. I built a prototype site where developers could try out our proposal for shared chat components, including switching designs and configuring features. The goal was to learn what other teams needed before engineers built the library.
Password required
04
SageMaker Canvas: Training models to classify images
Designed the workflow from uploading images to checking model results.
Worked with four engineering interns and a PM intern.
Business analysts needed a way to train models to recognize categories of images without relying on a data scientist. During my internship, I designed that workflow in SageMaker Canvas, with a focus on organizing and labeling batches of images before training a model.
Anagnori is an iPhone prototype based on practices that helped me in therapy. I use it to plan something I’d normally avoid, write down what I think will happen, then compare that with what happened. I’m the only person who has used it so far, and it hasn’t been clinically evaluated.
Fifteen film and print effects — burns, filmstock grades, crumpled paper, neon, a street light — running as real CSS/SVG/WebGL over live DOM instead of baked into a render.
A running-readiness app I took from a Jupyter notebook to a deployed private beta in about twelve days, leaning hard on AI agents. Built multi-user; so far I'm the only one on it.
