Resume

Amazon

• Build the agentic design tooling ~40 teams use to ship one shopping surface against Amazon design systems — generated from a prompt, not hand-coded.
• The public version of this problem is Luthier, three roles down. This work is internal, so the case study stays high level.

Void Industries

• Built a real-time collaborative game engine, four engineers, pre-launch.
• Real-time collaboration is the hard part of a design tool and the reason this four-month contract is on the page: it is the same problem as multiplayer editing in Composer, solved again from scratch with nobody else’s library to lean on.

Amazon

• Cut add-to-cart latency from 300ms to under 50ms with optimistic updates, and designed the interaction as a component every add-to-cart surface reuses.
• Shipped the first Rio Design Tokens, the token layer the Amazon app themes from.
• No designer was assigned to that interaction, so I designed it — the choreography, the easing curves and the timing — and reviewed it with a principal designer. The part a handed-over spec cannot carry is feasibility: what the motion should say and what holds up at frame rate on a busy page were the same decision, made once. It launched on the homepage and Buy Again.
• Rio Design Tokens is what makes the app themeable, consistent and accessible at once: one source of truth for colour, type and spacing across iOS, Android and web, instead of three teams agreeing by hand.

Amazon

Marketing graphics, Amazon worldwide. Lead FE, team of five.

• Took Composer from 5,000 to 50,000 graphics a year. Built Luthier, a no-code template editor, and a Figma plugin that turns a designer’s file into a template.
• Built Brand Builder: 100 templates per brand across 30+ brands, 3,000+ in all, against 20 a year hand-coded.
• Launched an A/B test increasing product image size on Amazon Fresh: +$85.4MM annualised, +0.67%.
• The rate is the whole argument: a hand-coded template took weeks and only its author could change it, so the ceiling was whatever one engineer could write in a year.
• Migrated the rendering engine from Canvas to HTML, making a new template 5× faster to create, and cut initial image load 75% (5s to 1s) by profiling loads and caching repeated calls.
• The Fresh test result held at 99.88% probability of positive return. Led ideation, requirements, prototyping, usability testing and both ends of the build.

Microsoft

• Shipped the PowerPoint Pen Toolbox after 20-plus prototype rounds in code, with an A/B test settling the final design. Shared components used by 100M+ people.

University of Washington

Johns Hopkins University & the Peabody Institute