things I've built

Real products, not demos.

Practical builds across AI agents, finance/life tracking, local-first media tools, VR, and production AI video, focused on shipped systems with clear user value.

theme shipped systems with clear user value range AI · SaaS · Web3 · VR · platforms

Selected work

1 / 5

Native macOS AI agent Active product

Chapeta

Instant AI everywhere on the Mac. One shortcut summons an agent that acts across apps, files, the terminal, and the screen — and drives a goal to completion while you stay in control.

Challenge: go past chat-in-a-box AI to a real agentic harness that takes a goal and accomplishes the mission on a live desktop.
Role: built the SwiftUI/AppKit menu-bar app end to end — model routing across 100+ models, the agentic loop, tool orchestration (files, shell, AppleScript, screenshots, computer use), subagents, and graduated approvals.
Result: a local-first agent spanning supervised to full autopilot, completing multi-step missions behind one keystroke.

SwiftUI/AppKit
Agentic loop + tool orchestration
Computer use & subagents
Local-first, graduated approvals

1 / 5

AI life-tracking SaaS Shipped product

Gepeta

An AI life-tracking SaaS that turns messy personal inputs into structured finance, health, and productivity records users can search, analyze, and act on.

Challenge: make personal tracking effortless enough to become a daily habit.
Role: built the AI extraction and routing core, then shaped the product flows around text, voice, images, spreadsheets, privacy, and dashboards.
Result: a cross-platform product that structures records across 13 life aspects from text, voice, photos, and spreadsheets.

AI data extraction
13 active life aspects
DynamoDB + typed schemas
PWA + Capacitor iOS

1 / 4

Last-mile media editor Shipped product

FastEdit

A local-first last-mile media editor for the file fixes that happen right before posting, sending, or publishing — convert, compress, trim, crop, redact, and export without uploading private files.

Challenge: help users finish the small but blocking media tasks that usually send them hunting across upload-based tools.
Role: built the local editing engine for stills, animations, and short video, with fit-to-size export, presets, timeline tools, overlays, filters, blur/redaction, and batch flows.
Result: a shipped privacy-first tool supporting 15 input families, 12 export formats, and 79 platform presets.

Last-mile media fixes
100% client-side processing
15 inputs / 12 exports
79 platform presets

1 / 4

VR game systems Reality Shift

Terragon: Symbol Of Magic

A VR game systems build that combined real-time interaction design, procedural gameplay, enemy behavior, and ML-backed gesture recognition.

Challenge: make hand-drawn VR gestures reliable enough to feel like part of the core combat loop.
Role: built the gesture-recognition flow from raw 3D coordinate normalization through dataset preparation, CNN training, and runtime detection.
Result: shipped and sold a fully functional VR game with core systems across input, interactions, combat, procedural levels, enemy AI, animation timing, and audio sync.

Unity + OpenXR
CNN gestures
3D normalization
Procedural AI levels

AI research automation Q.E.D Science

Editorial Grant Agent

An AI research automation workflow that turns grant discovery, eligibility checks, evidence gathering, and drafting into a repeatable product flow.

Challenge: convert a high-context editorial research process into software that can reason, gather evidence, and draft reliably.
Role: owned the full product pipe across frontend workflows, backend orchestration, data flow, and the tool-backed AI engine.
Result: repeatable research operations for discovery, fit checks, evidence collection, drafting, review, and application preparation.

End-to-end product pipe
AI engine + tools
Eligibility reasoning
Draft automation

1 / 3

AI video product engineering Professional work

D-ID Product Foundations

Production product engineering across D-ID Studio, Chat, Agents, and API-facing AI video surfaces used to create and integrate generative video workflows.

Challenge: build product foundations that could support multiple AI video surfaces as the platform expanded.
Role: led delivery and integration work across frontend architecture, API-driven workflows, Studio creation tools, Chat experiences, Agents, and API contributions.
Result: durable product foundations for core D-ID experiences that could evolve across creation, conversation, agents, and developer-facing workflows.

AI video product UX
React/TypeScript
API integration
Product architecture at scale