Failed projects.

Things I built up until 2025. Most of them work, none of them are polished, and none have users. What I've done since is on the homepage.

• Forage Meal-discovery app that checks your pantry and orders the missing ingredients for grocery curbside pickup. Swift, Expo, Spoonacular, Kroger and Instacart APIs, Stripe. Only got access to 3 of 6 APIs.
• Unvibecode SaaS that rebuilds a vibecoded site into production-grade output with the Claude API. React, Supabase, edge functions, Stripe, Vercel. Didn't work with more complex sites.
• Tidysheet CSV and spreadsheet cleaner shipped as a Google Sheets add-on on Cloudflare Workers. Working, but mediocre UI with mediocre model implementation.
• Landassist Screening and listing tool for small Georgia landlords, built on Next.js. Landlords prefer their existing systems.
• Mouse / Metamouse Hands-free cursor control. Gaze-driven macOS cursor with dwell-to-snap, then aim estimation from a Meta Glasses camera frame. MediaPipe, OpenCV, AprilTag registration, macOS Accessibility API. The glasses-to-cursor path had high latency.
• Beacon Locating missing people by matching gait across authorized camera feeds, built around an Amber Alert workflow. Python, pose estimation. Re-ID across cameras wasn't accurate enough, and consent had no clear answer.
• Polytrack AI Reinforcement-learning agent trained to drive the browser game Polytrack. Python, PPO. Never trained past the environment.
• Rosterup Youth sports team-management SaaS with a validation-first plan and a $1k budget. Free apps are better than sports management software could ever do.
• Bustem / Bustem2 Static HTML and Next.js rebuilds of the cal.com marketing homepage. Built as a rewrite of the site, but never implemented.
• Inavolu Earlier personal-site build on HTML, CSS, and Netlify. Replaced by this site.