Personal repair for AI-built apps
I fix it and prove it works.
Send me the app you're stuck on. I'll tell you what's really wrong, free.
Built with Lovable, Bolt.new, Cursor, v0, Replit, Claude, or ChatGPT? You're in the right place.
Noumenon / App review
Demonstration on a test app
You've asked it to fix the same thing five times.
It says it works. You click. It doesn't.
You shouldn't have to guess what to trust.
How it works
A repo link, an export, or the app URL plus a test login. Start with what you have. I'll help you with access.
I run it from a clean checkout, a fresh copy of your code. I tell you exactly what's really broken and what the AI faked.
Agree on the price first. I fix it and hand you a report that proves it runs, with the steps and results you can check yourself.
Why this is different
Anyone can tell you it's fixed. Your report shows what I tried, what actually happened, and how to repeat the check.
You see which parts work, which parts fail, and which parts only pretend to work. After the repair, you get the new results, plus anything still untested or unresolved.
No need to take my word for it.
Read the sample diagnosisWhat I fix
Studio records
One on an open-source project, the rest on the studio's own live apps. None is a client project yet.
Broke: a message with a file mention could not be sent when the file path contained a space.
Found: reproduced it, traced it to the mention scanner stopping at the first space, and wrote a regression test that failed before the fix and passed after.
Fixed: fix and test submitted as pull request 3925. The maintainer had landed his own fix minutes earlier, closed the pull request and wrote: “Thanks for the fix and the regression test.” The method is the point: same bug, same root cause, proof attached.
Broke: desktop navigation was invisible on the live page.
Found: by looking at the page, after it had passed review without anyone looking.
Fixed: restored the desktop navigation.
Broke: a production outage.
Found: the install flag --omit=optional caused it.
Fixed: never install without optional dependencies.
Broke: an empty password showed a generic “could not complete that request” message while the backend was healthy.
Found: App Review rejected version 1.0 because it read as a server failure.
Fixed: field-level messages in build 11, live since this date.
Broke: subscription review stalled.
Found: a hidden MISSING_METADATA state turned out to be territory availability.
Fixed: set availability through the API for 175 territories.
Pricing
A few questions
Yes, to run and repair the app I'll need access to the code. You can share a repo or an export. If you only have a live link, start there and I'll help you find what I need.
I use it only to diagnose and fix your app. I don't share it, sell it or publish it, and if you'd like an NDA, I'll sign a reasonable one before you send anything. Within 14 days after the job ends, I delete my copies and you can remove my access.
Yes, as one of my tools, with settings that stop it from training on your code. The difference is what happens next: nothing is called fixed until the checks pass, and you get the results to check yourself. If your code can't go through any AI tool, tell me first and I'll quote the work without them.
I reply within one business day, and the free diagnosis usually takes two business days once I have access. For paid work, the timeframe is in the quote, before you agree to anything.
The diagnosis is free. For paid work, you get a written quote first and pay by card through a secure link when you accept it. Jobs over 1,000 USD are half up front and half on delivery. If the agreed checks don't pass and I can't make them pass, you get a full refund.
The diagnosis looks at the one problem you describe: I reproduce it, find the cause, and tell you what fixing it would cost. The review covers your whole app, including login, who can see which data, payments, exposed keys and what breaks when it goes live. You get a report ranked by what to fix first, and if you book a fix within 30 days, the 149 USD counts toward it.
If a problem I fixed comes back within 30 days in the code as I delivered it, I fix it again for free. If a later change broke it, whether from you or your AI tool, I'll quote before doing anything. The proof report makes it easy to tell which happened.
I'll tell you what's blocking it and what your options are. The diagnosis is free either way, and you only pay for work you've agreed to in writing.
No. Tell me what you want the app to do and what happens when you try. I'll explain the findings in plain words and help you share the files.
A link to the app or repo and a description of what's wrong. An export works too. If I need a test login, I'll ask by email. Please don't send your own password or secret keys through the form.
About
I build with these same AI tools every day. I kept hitting the wall you're probably at now: the AI says it's fixed, and it isn't.
So I spent the last year building tools that check the work instead of taking the AI's word for it. You can read the public code yourself: AutoMaxFix, hexis-mcp-guard and cve-guard.
Every job runs the same way. I reproduce the problem on a clean copy, fix it, then run the checks again and show you the results. It isn't done until they pass.
There's no team and no handoff. The person who reads your message is the person who fixes your app.
Let's look at it
You don't need to explain it perfectly. Tell me what you expected and what happened instead.
I'll reply personally by email.
The diagnosis is free. No commitment.