Pulse

Monitors 9 systems (bots, sites, pipelines) via a mix of push heartbeats and pull HTTP checks. Sentinel detects state transitions (up→down, down→up) with symmetric hysteresis to kill flapping. A daily Telegram briefing arrives every morning via Apollo-Pulse. The dashboard shows the live state of my entire deployed infrastructure at a glance.

highlights

• +Push + pull hybrid: bots fire heartbeats, sites get polled by pg_cron every 5 min
• +Sentinel: hysteresis-based alerting (2 unhealthy to alert, 2 healthy to recover)
• +Daily briefing bot — dead-man switch; silence = Vercel is down
• +59 tests across sentinel logic, display, and data transforms

what was hard

Off the request path: zero writes on dashboard load (pg_cron owns all polling); Sentinel dedup: only 1 alert per transition regardless of how many poll cycles