GitKaz
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

AURA helps small teams keep their Python applications running without a dedicated operations team. It connects to Docker, watches application health, and investigates failures using logs, resource measurements and configuration history. The owner decides which recovery actions are allowed. Every action is followed by checks against the running application. In our live demo, an incorrect port makes a real Linkding instance unreachable. AURA identifies the mismatch, restores the saved configuration and verifies that HTTP requests work again. One recorded cloud run took 97.8 seconds and three model calls, with no human intervention after the fault was introduced. A separate code-repair workflow tests proposed changes in an isolated container and requires approval before deployment. We used it on a historical Linkding file-download defect: the AI proposal needed a human correction, then both 500 MiB downloads completed with matching hashes under a 256 MiB memory limit. The interface keeps the investigation readable, with a particle portrait reflecting its progress and detailed evidence available when needed. AURA records restored availability separately from a verified code fix, so users can see exactly what was achieved.