Building Patch — incident investigation in Bubble

Something broke. Patch tells you what probably caused it.

I’m building a tool that pulls evidence from three places developers already use and correlates it into one timeline:

  • Sentry — what broke
  • Vercel — what deployed
  • GitHub — what changed

Then it ranks the evidence and asks an LLM to explain the most likely cause, with a confidence score, the reasoning behind it, and a recommended next step.

Example: a payment API starts throwing 500s at 14:08. A deployment went out at 14:03. The commit behind that deployment touched the exact file in the stack trace. Patch surfaces that chain instead of making you open three dashboards.

A few things I’m figuring out along the way:

  • Doing the correlation scoring without backend workflows. Currently leaning on Toolbox to run the ranking logic and pass results back into Bubble.
  • Keeping the confidence score honest. It comes from deterministic scoring rules, not from the model inventing a number.
  • Making a demo that opens with zero setup, so nobody has to connect their own Sentry account to see it work.

I’m also running a small real app on Vercel with a deliberate bug in it, so the demo data is a genuine incident rather than seeded records.

Still mid-build. Happy to hear thoughts, especially from anyone who’s pushed correlation-style logic through the API Connector.