Anveeksha on Anveeksha

Anveeksha connects Jira, Confluence, GitHub and Slack into one event stream, resolves who did what across tools, and detects what has shifted against each project’s own baseline. The most honest place to test that is our own delivery, so the first workspace the product ever ingested was ours.

It also keeps the practice and the product in one loop. The engineers who deploy AI for clients are the same people whose tickets, pull requests and conversations the system is reading.

A serialisation fault had silently stopped a project’s baseline from being applied, so nothing could register as a shift. The system reported the silence itself: activity was arriving, but no variance had been computed for longer than the baseline allowed.

Replaying history as it was lived exposed a timezone bug across 14 repositories. Events lined up wrongly around day boundaries and appeared as a surge nobody on the team recognised. The receipts showed the timestamps were wrong, not the people.

A malformed webhook arrived in production. The ingestion layer quarantined it and carried on rather than dropping it or stalling the stream, and the event was replayed once the payload was understood.

In each case detection was deterministic. Thresholds, severity and state were computed, not generated. The language model wrote the explanation, with the evidence it was computed from one click away; it never set the severity.

That is the same standard we build to for clients: every claim carries its evidence, and you can inspect it.