I design AI systems that power creative production.

Problem: Artists needed a fast, native tool for concept art that integrates AI editing without leaving their canvas. Built: A full macOS drawing app with AI-powered editing, layers, brushes, and a Procreate-inspired UI, designed in collaboration with the art team to fit real production workflows.

Problem: Patients tapering off medications need a safe, clear way to track gradual dose reductions, while providers need a way to share evidence-based taper plans. Built: A native iOS app with separate patient and provider modes, on-device data privacy, daily dose reminders, and QR code plan sharing between clinicians and patients. Shipped end-to-end and live on the App Store.

Problem: Converting video sequences to sprite sheets took artists hours of manual work per asset. Built: A production pipeline with AI-powered grid detection, chroma key, and multi-format export. Integrated into the team's existing workflow, reducing a multi-hour process to minutes.

Problem: Motion designers constantly context-switched between AE and external AI tools. Built: A 10-module AI extension embedded directly in After Effects: image expansion, upscaling, background removal, video generation, audio, and copywriting. Developed with the motion team to eliminate round-trips and keep artists in their native environment.

Problem: Casual mobile puzzlers tend to be either mindless tappers or punishingly tight - I wanted a cozy, satisfying merge-and-build loop that rewards spatial thinking without stress. Built: A voxel-sculpture merge puzzle as a fully playable PWA - drop blocks to fill ghosted 3D shapes, with daily challenges, a gallery, and a chill "Zen" mode. Designed and built end-to-end: game design, art direction, and code.

Problem: The creative team needed one unified interface for multiple GenAI capabilities instead of juggling separate tools. Built: A modular production platform for image generation, video processing, 3D modeling, and background removal, on Replicate API, deployed for daily use by the creative and production teams.

Problem: Existing YouTube downloaders are either sketchy ad-laden web services or command-line tools nobody non-technical wants to touch. Built: A native desktop app — SwiftUI on macOS, a Flask + Edge --app shell on Windows — wrapping yt-dlp, ffmpeg, and deno into one drop-in installer. Quality picker (up to 4K), trim, multi-language subtitles with auto-translate, "Make compatible" mp4 re-encoder, inline player, self-updating yt-dlp. Shipped as signed .dmg and Inno Setup installer with GitHub Actions auto-builds.

Problem: Burning styled captions into videos is a slow round-trip through Premiere or paid SaaS, and YouTube's auto-translate is famously out of sync. Built: A self-hosted FastAPI app that runs OpenAI Whisper on the audio, optionally polishes the segments with Claude, and burns the result through FFmpeg + libass. In-browser editor with a timeline, drag-to-position overlays, and a live preview that matches the burned output to the pixel. Bundled into the YouTube Downloader's "Caption" tab via an embedded WKWebView.