- An iPad app on a mission to make kids' screen time worth it: activities that use voice, computer vision and generative models to send the kid back into the physical world.
- Working prototype — the draw activity: the agent suggests what to draw, generates an instant reference image, starts the timer, and the kid draws on paper, off the screen; the agent then helps photograph the drawing and keeps the artifact. Five-plus minutes off the screen, an artifact at the end. Play-tested with my own kids.
Voice, computer vision, fast image models, live camera
- One selfie plus any reference image becomes a full photoshoot of you. Idea to App Store in one month; defined, designed, built and launched solo.
- Server-driven model registry: every model and its generation function live in Supabase, so a newly released image or video model reaches users the same day it ships — no app update, no App Store review. 20+ image and 10+ video models wired in; turnaround from a trending model to live in the app is hours.
- Owned distribution and monetisation as product: ~2M views from a single UGC creator channel, ~10,000 downloads, 4.7 rating, 2–3× conversion lift from paywall and pricing experiments, 8.43% paid conversion, 5,206 new customers, peaking at 254 active subscriptions and $2,170 MRR.
React Native / Expo, Supabase, fal.ai, Replicate, Gemini
meracl.app — creative intelligence for mobile apps
- Finds what is winning on mobile: scans competitor organic and paid creatives across Meta, TikTok and Instagram, discovers competitors by ad destination rather than brand name, and ranks creatives by engagement signal.
- Generates ad variants that reuse a winning format with hooks rewritten for your app, ready to run the same day. Example scans show +38% to +55% CTR against the competitor original.
- In use by Endel, Goodie, Stelth and Aya Photo. Daily board refresh flags new pages and new high-performing creatives automatically.
App store and social scraping at scale, first-frame fingerprinting, generative video and image models
aigma.co — capability benchmark, then a tool
- Benchmarked every frontier model on how precisely it rebuilds an image or screenshot as working HTML — a capability question no public benchmark covered. Gemini won on layout fidelity.
- Turned the finding into a product: reference image in, real website out, on an infinite canvas. Prompt work aimed specifically at escaping the AI-slop look.
- Shipped a CLI so Aigma plugs into terminal coding agents and generates design options from the command line. Roughly 200 registered users.
HTML, canvas UI, frontier model APIs, CLI
- A minimal, customizable journaling app with Ezra, an AI companion for reflection, insights and guided entries. Designed and built solo, from a daily journaling habit and a frustration with noisy journaling apps.
- Complete cycle owned end to end: concept, design system, build, App Store submission, subscriptions.
React Native / Expo, Supabase, LLM APIs
Oval AI Skin Analyzer · Cellr Cell Tower Map · Watumean2025–2026
- Two further shipped iOS apps across beauty-tech and utilities, and a WhatsApp agent experiment — eight personas behind one number, with per-user memory and an ad-to-conversation-depth funnel.
React Native / Expo, Supabase edge functions, WhatsApp Business API
