God's Eye

Here’s a more natural, dev-on-Twitter version:

Built Prometheus because I got tired of playing human router between GitHub, Jira, and Slack.

You know the drill:

A PR sits unreviewed for 3 days.

The Jira ticket is still "In Progress" because it depends on that PR.

CI is failing downstream.

Someone is asking in Slack why staging is broken.

And somehow, nobody connects all of that.

So we built Prometheus.

It pulls data from GitHub, Jira, and Slack, then figures out what's actually blocking the team.

For example:

PR #1 has been waiting 3 hours for review. It's blocking KAN-4, which is holding up downstream tests.

Then it ranks the bottleneck and drafts the message to the person who can unblock it.

The important part: it doesn't send anything on its own.

You get a draft and a 1-click "Approve & Send" button.

AI agents doing random write actions without asking sounds cool until your bot starts spamming your team.

A few things we learned the hard way:

Jira changed an API endpoint and we spent way too long debugging a 410 Gone.
We accidentally had mock data fall back into production, so a fake developer named Alex kept appearing in Cloud Run.
Posting in #general was both annoying and required extra Slack permissions, so we switched to sending DMs to the actual reviewer instead.

Stack is pretty boring:

Python + FastAPI + SQLite Gemini via google-genai / Vertex AI GitHub, Jira, and Slack APIs Cloud Run + Firestore Plain HTML + Tailwind + vanilla JS

No magic. Just trying to answer one question:

What's blocking us right now, and who can unblock it?

Biggest takeaway so far:

Don't make the AI act first.

Let it do the detective work, draft the action, and let a human hit approve.

Built With

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.