AI-Powered Operations Dashboard

The problem

Running several independent systems means checking several places for whether anything's actually broken. That's the classic "single pane of glass" problem — and it's exactly the kind of small internal tool that saves real time once it exists and gets ignored once it doesn't.

What I built

A dashboard that pulls health signals and key metrics from several independent systems — process status, database freshness, scheduled-job health — into a single, auto-refreshing view. The more interesting part: every morning, a scheduled job hands the day's collected status data to the Claude Code CLI itself, running under an explicit, scoped prompt — with hard rules like a word limit and a restriction to only write the one output file — and it writes a short, plain-English operations briefing that appears on the same dashboard.

System flow

Technical depth

• A pluggable collector architecture fuses multiple signal types — process state, database freshness, HTTP checks, log parsing — into one consistent health model.
• The daily briefing runs the Claude Code CLI as a scheduled automation step, not an interactive chat session, with explicit scope and output constraints rather than an open-ended prompt.
• Graceful degradation: a missing or stale status source shows as a clear warning state rather than crashing the dashboard.