Inspiration
When a support engineer picks up an incident, the first job is to separate the log evidence from a plausible explanation. Hybrid Incident Analyzer turns a sample into a working diagnosis, supporting lines, uncertainty, and two scoped checks.
How it works
Choose from disk pressure, DNS timeout, and untrusted input. The workspace defaults to Device only. Device first uses Firebase AI Logic to prefer Chrome’s local model, with App Check and explicit consent before a synthetic log may fall back to the cloud. The result shows the actual inference source, elapsed time, and a local JSON export.
What we learned
The local model and the cloud route have different prerequisites. We verified an offline response on Chrome’s Gemini Nano 4 and a separate cloud fallback with App Check and explicit consent. The app reports the actual source returned by Firebase. Logs are treated as data and the app never executes a check.
New work
Built during the contest window. See NEW_WORK.md in the project repository for scope and attribution.
