Thien Tran

University of Washington

June 2026

Bachelor of Science in Informatics: Human-Computer Interaction

Seattle, WA

• Designed and shipped a unified system for the platform's flagship infinite-canvas workspace, owning it from Storybook prototype to React code in a ~10k-file codebase, consolidating three features into one interaction model.
• Designed and implemented a prompt queuing system that lets users stack AI generation requests that auto-advance as each completes, shipped to production with 45 unit tests after restructuring the UI into a pinned composer dock.
• Initiated the design system's first motion token language after auditing the full codebase (53 ad-hoc durations, 9 spinner implementations, 0 tokens), collapsing them into a 6-step scale of CSS-var tokens with canonical loading and empty states.

• Designed SETLib, a searchable system organizing 17 years of CS curriculum, cutting facilitator prep time by 81% and projected to save $35K annually.
• Partnered with 5 engineers to map initial AI parser constraints, designing defensive UX with side-by-side review and manual correction tools to explicitly surface extraction failures from complex math and embedded images.
• Designed V2 confidence scoring framework for an upgraded hybrid LLM parser, reducing mandatory reviews by 64% while maintaining 96% content accuracy — driving 100% facilitator adoption after professors mandated the switch.

• Engineered an open-source visual editor for React across Vite, Next.js, and Turbopack that resolves DOM clicks to exact source locations via a custom Babel plugin, enabling direct AST rewrites of source code.
• Designed a direct-manipulation editing surface that gives designers the visual control of a design tool on the live product itself, turning every edit into a real, reviewable code change instead of a spec or a ticket.
• Built a fail-closed UX that refuses unsafe edits and drops contextual flags, routing precise JSON work-orders through a custom MCP server to the user's local AI agent for complex architectural changes.

• Designed a mobile-first behavioral trading app with a cross-functional team of 5, reframing missed trades in retail investors as material for self-awareness rather than execution analysis.
• Led the V2 redesign after MVP testing showed beginners were intimidated by data-heavy design, pivoting from an analytical product to an editorial journal that 10/10 testers preferred in comparison.
• Shipped and designed V2 using Claude Design, Figma MCP, and Claude Code, completing the full pivot inside one academic quarter on a cross-functional team of 5.

Design: Product Design • Interaction Design • Visual Design • Service Design • Information Architecture • Wireframing • Prototyping • Design Systems • User Research • Usability Testing • A/B Testing • Accessibility (WCAG 2.1)

Tools: Claude Design • Figma • Figma Make • Lovable • V0 • ProtoPie • Miro • FigJam • Adobe Photoshop

Development: HTML/CSS • React • Python • Next.js • Claude Code • Codex • Google Antigravity • SwiftUI