Send me one AI feature, a chatbot, a search or retrieval flow, an agent, an automation, and I'll tell you where it actually fails, not where the demo looks fine.

Most AI features are built to work once, in front of the person who built them. This audit looks at what happens the other 999 times: bad inputs, empty results, cost blowing up under real traffic, and data going somewhere it shouldn't.

You'll get a short written report covering:

The 3 most likely ways this feature breaks in production

Any data exposure or security issue I can see from the outside

A rough read on cost per call, if I can estimate it from what you share

The one fix I'd do first if it were mine

This is a diagnosis, not a fix. I'm not writing code or making changes, you walk away knowing exactly what's wrong and in what order to fix it. If you want the fixes built, that's a separate conversation once you've seen the report.

What I need from you: a link, a Loom, screen-recorded walkthrough, or read-only access to the relevant code/flow. No live credentials, no production database access.