LLMs provide an interesting opportunity to non-programmatically investigate the cause of a failure of an integration attempt.

Create a bot to listen for gerrit change-integration-failed events and feed the failed log, test case source (if relevant) and the change diff into an LLM. The results must be consistent and accurate enough to not annoy users.

The output can be posted to the change.