I fix broken AI-built apps.

Apps generated with Lovable, Bolt, v0, Replit, Cursor or Claude Code usually get to about 90% quickly. The last 10% is where they stall — and the failures are often silent, so the demo looks fine and the problem surfaces in production.

What I work on

• deploy and build failures (Vercel, Netlify, Replit)
• environment/config problems — "works locally, breaks live"
• Supabase auth and RLS
• broken forms, signup and checkout flows
• API and webhook integration failures
• silent data loss (writes that report success and never land)

Samples

ai-app-failure-samples — runnable reproductions of these bug classes, each with the broken version, the fix, and a regression test that fails on the bug and passes on the fix. Clearly labelled demonstrations, not customer code.

How I work

AI-assisted engineering, verified with automated regression tests and adversarial QA before delivery. Fixed scope, and I'll tell you when something is out of scope instead of guessing.

Need something fixed? Two ways in:

• Public problem, no secrets — open a Vibe-Code Rescue Request. I read every one and reply with the likely cause and the next diagnostic worth running.
• Private project — use the Everstead App Rescue intake, which also offers a free three-blocker teardown. Please don't paste credentials or private source into a public issue.

Scope and pricing are written down in SERVICES.md (fixed-scope work typically starts around $30 when the problem is genuinely small). There's also a separate paid diagnostic — the $99 Mini Launch Readiness Review, an AI-assisted one-page readiness report with no human review included.