I build intelligent systems
that do real work.

Objective

Create a storefront brand with a distinct red-team identity — sharp, confident, unmistakably not a template.

Problem

Security merch lives in generic Shopify themes. The audience is design-literate and allergic to default.

Architecture

A fully built HTML prototype to lock the aesthetic, then a Shopify Hydrogen implementation blueprint for a headless, fast storefront.

Tech stack

Challenges

• A bold aesthetic that stays shoppable
• Translating a static prototype into Hydrogen cleanly
• Performance under heavy motion

Solution

A tight token system and GPU-friendly motion, so the cyberpunk feel survives the jump to a real commerce stack.

Key features

• Signature override tagline system
• Headless, fast storefront plan
• Reusable brand tokens

Future

Live product drops, account tiers, and a member-only arsenal.

Objective

Turn a desktop arcade prototype into a polished, shippable Android game.

Problem

Desktop controls, layout, and pacing don't survive a small touchscreen without real rework.

Architecture

A canvas game core wrapped in a full Android Studio project, re-tuned for touch input, difficulty selection, and a condensed mobile HUD.

Tech stack

Challenges

• Touch controls that feel precise, not floaty
• Normalizing the "heat" mechanic for mobile
• HUD that fits without crowding play

Solution

Reworked input handling, a tuned difficulty selector, and a condensed HUD — packaged for Play Store submission.

Key features

• Touch-native controls
• Difficulty selection
• Play-Store-ready build

Future

Online leaderboards and a daily challenge mode.

Objective

Build a production-grade AI code reviewer — one that reasons like an expert engineer, not a linter. Returns typed findings with severity, evidence, and corrected code.

Problem

Static analysis tools catch syntax errors. They miss SQL injection flowing across function boundaries, N+1 query patterns, weak cryptography choices, and architectural debt — the issues that actually matter in code review.

Architecture

React frontend with a schema-locked Claude Sonnet backend. A strict 9-field JSON output contract enforces structured findings; every result is validated before rendering. The model sees the entire file in a single pass — not line by line — to catch cross-function issues.

Tech stack

Challenges

• Preventing hallucinated findings via schema enforcement
• Designing confidence scores that mean something
• Mapping findings to CWE/OWASP without over-claiming
• Building a text hierarchy that doesn't collapse under dense output

Solution

A schema-locked system prompt with four reviewer personas (Senior Engineer, Security Expert, Perf Engineer, Architect), per-finding confidence scoring, and a validated JSON extraction pipeline that gracefully rejects malformed responses.

Key features

• Four AI reviewer personas with distinct system prompts
• Six quality scores: Security, Maintainability, Readability, Performance, Testability, Overall
• CWE / OWASP references per security finding
• Collapsible analysis pipeline explainer
• Markdown report export

Future

GitHub PR webhook integration, CI/CD pass/fail status checks, and multi-file repo analysis via RAG over the codebase.

Objective

Build a visually stunning, interactive tool that teaches spatial dimensions from 0D to 7D — where the interface itself feels like it exists in higher-dimensional space. Every dimension gets its own live geometry, not just a diagram.

Problem

Most dimension visualizations are static diagrams or require heavy WebGL frameworks. There's no interactive tool that lets you explore all eight dimensions with real math, smooth motion, and explanations that are both technically accurate and genuinely interesting to read.

Architecture

Three-layer canvas stack: a background layer (stars, holographic perspective grid, ambient particles), a foreground layer (dimension geometry rendered per frame), and a pointer-events UI overlay (glassmorphism panels, mobile drawer, comparison modal). All 3D projection math is hand-rolled. No dependencies.

Tech stack

Challenges

• Cascading 4D→3D→2D projection for hypercubes (tesseract to hepteract)
• Momentum inertia: drag velocity bleeds into rotation after release
• Two-canvas compositing without a WebGL context
• Mobile bottom-drawer UX replacing the floating desktop panels
• Per-vertex depth-sorted edge rendering for 7D (128 vertices, 448 edges)

How 4D–7D rendering works

Each n-cube is generated combinatorially: 2ⁿ vertices, edges connecting vertices that differ by exactly one bit. Rotation matrices are applied in pairs of axes (xw, yw, zw, vw…) then projected through perspective division — a cascade of shadows, each dimension one step removed from true perception.

Key features

• 8 dimensions with real-time interactive geometry
• Drag to rotate · scroll/pinch to zoom · double-click to reset
• Dimension comparison modal (name, coordinates, directions, boundary, experience)
• Glassmorphism axes & controls panel + floating info panel
• WebAudio click and dimension-change sound design
• Full mobile layout with bottom drawer (Axes / Controls / Info tabs)