My AI-Augmented
Engineering Stack
My process is evolving, shaped by what I learn as I work across different phases and as new products or tech come online.
(Claude Design isn't making the list so far.)
My process is evolving, shaped by what I learn as I work across different phases and as new products or tech come online.
(Claude Design isn't making the list so far.)
Type any experience — catching a whiff of coffee, locking eyes with a stranger, stepping barefoot into wet grass — and watch the brain light up in response. Each region activates step by step with an explanation of why it fires and what role it plays, all inside a low-poly head you can rotate and explore.
Turn on face tracking and the head becomes a puppet that follows your movement, with wireframe hands that mirror yours. That part's experimental and just for fun.
Modifiers like fatigue and stimulants are on the way to show how the same moment plays out under different conditions.
Anthropic API (Claude Sonnet) Custom GLSL Shaders Google Apps Script MediaPipe Next.js Three.js Vercel
A gaze-controlled arcade game for macOS inspired by Atari's Missile Command. Players defend cities leveraging built-in macOS Accessibility — Eye Tracking or Head Pointer to aim, with Dwell Control or Facial Gestures to fire. Visual style inspired by classic Atari arcade art and EC Comics.
ChatGPT Claude GitHub Kling macOS Accessibility Features SpriteKit Swift Xcode
Each profile is built from primary sources and biography — mapping the formations that produced them, the gap between the person and the persona, and the psychology underneath both. Carl Jung called that gap the shadow.
These profiles let the shadow speak. Each one is psychological scaffolding — tone, vocabulary, self-awareness, the topics that open and the ones that close — assembled from the record. Future iterations will deepen the voice with documented cadence from each subject's own interviews and writing.
Bring a question, a current event, or just start somewhere. The responses come from the psychological model, not the public record.
Anthropic API (Opus, Sonnet, Haiku) Google Apps Script Vercel
Retailers use algorithms to make every price look like a deal. "Was $25, now $22!" But was it ever really $25? TruePrice tracks actual price history so you know what products truly cost over time.
This matters most for people who buy the same items regularly — especially anyone managing a chronic condition. Once you can see a year of real prices, you stop guessing and start buying when prices are actually low.
Next up — training models to identify the patterns behind the pricing. Transparency into when discounts are real, when they're staged, and when the same product gets quietly marked up before being marked back down.
Chart.js ChatGPT Claude Claude for Chrome Cron Python PythonAnywhere Selenium SQLite vØ by Vercel VS Code
AI-powered prototype that recognizes skincare products in real time, classifies formulas by function, and aligns them with personal skin needs—bridging science, design, and everyday experience.
ChatGPT vØ by Vercel
A sparring app for arguments. Pick a topic, choose your side, and trade turns with Claude as both opponent and coach.
Every argument gets scored on clarity, logic, evidence, and persuasiveness, annotated inline like an academic text, audited for fallacies, and rewritten using different rhetorical strategies. A running scoreboard tracks who's winning.
Anthropic API Claude Design Next.js Tailwind Vercel
A ground station receiving Low Rate Picture Transmission (LRPT) data from Russian polar-orbiting weather satellites as they pass overhead. Images show a continuous strip of Earth captured during each ~10 minute pass.
Current Satellites: Meteor-M N°2-3 (137.1 MHz) and Meteor-M N°2-4 (137.9 MHz). Typical coverage: 2-3 visible passes per day per satellite over Minneapolis-St. Paul area. (Built off of a previous project, creating an aircraft location receiver.)
Claude GitHub Gpredict Python Raspberry Pi RTL-SDR SatDump SQLite
An AI-powered project that takes the same 3D building data behind Google Maps and transforms it into something entirely different — a stylized, scrollable isometric map of downtown Atlanta. The renderer is built and capturing the city.
Next, a fine-tuned image model will convert those photorealistic tiles into illustrated styles, with the architecture designed so new visual treatments can be swapped in by simply retraining the model.
Claude Code GitHub Google Maps 3D Tiles API MagicaVoxel Three.js VoxCity
ML notebook prototype that analyzes a user's language for specific behavioral signals — emotional cues, cognitive distortions, resilience markers, and motivational patterns. Each signal is detected by a specialized classifier agent, turning open conversation into structured data the system can act on: how to pace the conversation, when to offer more support, when to acknowledge growth.
ChatGPT Deep Research Kaggle ML Notebooks
ML Notebook showing how those signals get used. Instead of responding to the surface of what someone said, the system assembles a reply from the underlying pattern — the emotional intensity, the coping style, where the person is in their story, what kind of resilience they're showing.
Each response combines four things: empathic reflection, validation, cognitive reframing, and forward guidance. The result is a CBT-aligned reply built from behavioral evidence rather than keywords.
ChatGPT Deep Research Kaggle ML Notebooks
MyDailyEpic transforms daily struggles into epic tales. Share a frustration — missed bus, work conflict, life challenge — and watch it reimagined through fantasy, sci-fi, or mythology.
Grounded in cognitive behavioral therapy, the experience uses narrative reframing to help users identify core values and build resilience — no clinical language required. Inspired by Joseph Campbell's Hero's Journey.
ChatGPT Deep Research GitHub React Tailwind CSS VS Code vØ by Vercel
A fully client-side image editor built on the Canvas API, with layers, blend modes, a filter panel, and core tools — brush, fill, lasso selection, crop. Everything runs in the browser. No server, no dependencies.
Built in vanilla JavaScript to see how far a creative tool can be pushed without a framework. AI-assisted development let me move faster on the deeper features.
Canvas API GitHub Pages Vanilla JS
An accessibility evaluator that covers all 86 WCAG 2.2 success criteria — across all four principles: perceivable, operable, understandable, robust. Most automated tools catch around 30% of issues, mostly contrast and alt text. This one is built to find the rest.
Submit a URL, a screenshot, or code. You get back the testing method for each criterion, who on the team owns the fix (design, development, or product), the specific steps to remediate, and how the issue actually affects users.
Anthropic API Claude Code Netlify React Tailwind CSS W3C Standards
Most design reviews rely on gut feel or a single checklist. UX Audit Lens runs your interface through multiple UX frameworks at once — paste a URL, upload a screenshot, or drop in code — and returns findings tagged to specific principles from Nielsen's heuristics, WCAG, and cognitive load research.
Each finding tells you not just what needs attention, but which lens caught it.
The system uses retrieval-augmented generation (RAG) to ground its analysis in a curated UX knowledge base rather than relying on the model's general training.
Anthropic API React Tailwind VS Code
A conversational triage system trained on 25 years of Geek Squad call transcripts to guide customers through hardware and software troubleshooting. Combined decision tree logic, machine learning classification, and natural language patterns to reduce support call volume.
Started with a laptop MVP and expanded to all Best Buy hardware and software. The approach became the foundation for Best Buy's current support diagnostic system.
Its next design challenge: balancing customer control with guided problem-solving, replacing linear FAQs with a dialogue that adapts to the terms the user says or understands.
An AI strategy for Vetavize, a benefits app for veterans navigating the VA claims process. The roadmap progressed from basic writing help toward systems that could evaluate claim quality and track outcomes — each phase designed to capture which suggestions correlated with successful claims.
That data would feed VASPER AI, a model trained on what actually works with VA reviewers rather than what sounds right.
The projects above are recent AI experiments. My full portfolio covers 10+ years of UX strategy, product design, and research across healthcare, retail, enterprise, and Fortune 500 clients.