Founder

Canary AI

Hear the warning before it becomes an incident. Canary reads a team's Slack threads, call transcripts and post-mortems, and finds the risks hiding in them.

canary.pnoia.dev

Live, with self-serve sign-up. A real-world pilot is designed but not run yet.

Canary AI, in motion.

The facts

A quote for every risk
Each risk comes with the exact words it was found in. Code, not the AI, checks that the quote is really in the document. A risk without one is rejected.
1 to 25
Likelihood times impact, scored on a published rubric that follows ISO 31000 and PMBOK. Code does the maths, so anyone can recompute it.
A person decides
Every suggestion waits in a triage queue. Nothing enters the risk register without a human yes.
Secrets out first
Passwords, API keys and tokens are blacked out before anything is saved or sent to the AI.
4 days
From the first commit on 18 Oct 2026 to live on the internet, with real sign-up, on 22 Oct.

My part

Founder and product owner. Canary rebuilds a hackathon-winning prototype, point by point from the judges' written feedback.

I set the direction and made the calls; it was built with AI coding agents.

From the film

A proposed risk in the triage queue, with its cause, event and effect — Each risk, with the exact words behind it.

The quotes highlighted in the source chat log — Checked against the source, by code.

Likelihood 4, impact 5, score 20, critical — Scored on a published rubric.

The risk heatmap — The register, as a heatmap.

Canary sign-in page: hear the warning before it becomes an incident — The sign-in page at canary.pnoia.dev.

The documents in the film are Canary's made-up sample files.