Production-grade web products where interface quality, browser behavior, and real implementation detail matter. Work at the intersection of frontend engineering, browser extension architecture, AI-assisted systems, and cross-platform debugging — especially when a polished UI still has to survive real device constraints, service worker lifecycles, and production deployment realities.
- Browser-platform engineering: Chrome Manifest V3 extensions, service-worker-driven workflows, isolated UI injection, and frontend systems that have to work inside real browser constraints.
- AI-integrated product systems: Models as part of the product stack where they support workflow quality, reasoning, and UX — not as a substitute for real implementation or human evidence.
- Cross-platform debugging: Solving browser- and device-specific issues on physical hardware using Safari Web Inspector, ngrok, and live debugging workflows across iPhone, Safari, WebGL, and media/runtime edge cases.
- Release-minded frontend systems: Shipping with TypeScript, testing, CI, and production behavior in mind — treating maintainability, debuggability, and deployment quality as part of the product.
| Domain | Technologies | Focus |
|---|---|---|
| Frontend | React, Next.js, TypeScript, Tailwind CSS | Component systems, state flow, performance-minded architecture |
| Backend & Data | Node.js, Express, MongoDB, Firebase Auth | REST APIs, authentication, data modeling |
| Browser Platform | Chrome MV3, Service Workers, Shadow DOM, MutationObserver, Web APIs | Extension lifecycle, DOM orchestration, isolated UI injection |
| AI & Media | Gemini, OpenAI, MediaPipe, Google Cloud TTS | Product workflows, multimodal analysis, voice features |
| Mobile & Debugging | Safari Web Inspector, ngrok, iPhone camera handling | Real-device debugging, mobile QA, runtime troubleshooting |
| Ops & Quality | Vitest, Jest, GitHub Actions, Husky | Automated tests, CI discipline, release-minded workflows |
SigSent - The Human Signal Lab
A full-stack pre-send email testing product. Users create email variants, generate tokenized reviewer links, capture human-response signals, compare outcomes, and refine copy before launch — with a Gemini-powered assistant grounded in real campaign data rather than generic completions.
The product is structured around human evidence, not AI-generated predictions. It is a structured experimentation workflow for measuring trust, clarity, and action intent before an email is sent.
Stack: React, TypeScript, Tailwind CSS, Node.js, Express, MongoDB, Firebase Auth, Gemini API, Google Cloud TTS, MediaPipe
Architecture:
- Signal Lab workflow: Tests, variants, reviewer sessions, tokenized public review links, behavioral event capture, survey collection, and comparison reporting — structured as a cohesive product, not isolated features
- Public reviewer sessions: Secure token handling, expiration, stateful session tracking, and event capture without requiring reviewer authentication
- On-device processing: Optional MediaPipe FaceLandmarker for client-side facial-reaction capture within the broader workspace
- Cross-platform constraints: Browser- and device-specific work across Safari, iPhone, and client-side runtime handling, including multimodal input
CI/CD: GitHub Actions — lint, typecheck, tests, and build on every PR.
Status: Live product, actively developed · source is private
Portland Timbers Matchday — Published Chrome Extension with Safari Support
A Manifest V3 extension that shows upcoming Portland Timbers matches with a live countdown, TV/streaming info, and a fan confidence poll.
The Architecture: chrome.storage.local for state, alarm-based hourly refresh, runtime messaging between popup and service worker, and a three-tier data resolution strategy (live ESPN API → cache → bundled fallback). Safari support is handled through Xcode conversion with no polyfills required.
The Result: Published on the Chrome Web Store, with Safari support built into the codebase and workflow. GA4 telemetry via Measurement Protocol, Jest coverage, GitHub Actions CI, and Codecov integration.
Search Cashback Injector — MV3 DOM Injection Proof-of-Concept (In Development)
A Manifest V3 Chrome extension that detects supported merchant domains in Google Search results and injects inline cashback banners next to matching results.
The Architecture: TypeScript + Vite build. Content script handles result link detection and domain normalization; background service worker resolves offers via a configuration-driven merchant registry with chrome.storage.local caching and JSON fallback data. Injected UI uses Shadow DOM for full style isolation from the host page.
The Result: A focused proof-of-concept for safe DOM orchestration, content script ↔ service worker messaging, and isolated UI injection in live search environments. Vitest coverage with GitHub Actions CI.
OfferEngine — Zero-Backend Coupon Utility
A client-side coupon generator for configurable discount codes, QR output, and persistent storage with no backend required.
The Architecture: Next.js 15 (App Router) + TypeScript + Tailwind. Coupon codes use crypto.getRandomValues for suffix generation. QR rendering via qrcode.react. localStorage persistence across sessions. Input validation for discount ranges, code length, and expiration dates.
The Result: Deployed on Vercel with Vitest unit coverage, ESLint, TypeScript type-checking, and GitHub Actions CI.
Oregon Lawn Barbers — Client Site
End-to-end design and build for a Portland-area landscaping company. Conversion-focused, responsive, and deployed on Vercel with Next.js, TypeScript, and Tailwind.
I treat AI tooling as an architectural constraint, not a substitute for engineering judgment.
That means:
- generated code should fit the actual system, not generic defaults
- repo-specific rules and conventions should keep implementation aligned with the product
- core projects should validate through linting, tests, and CI where the workflow calls for it
- browser behavior, runtime detail, and production reality matter just as much as clean code in the editor
I care about understanding what is actually running — especially when the debugging gets messy.