(CASE STUDY · 01)

PULSE

Service Status
Official Feeds
React
Electron
Android

Pulse reads the official status feeds of 77 services developers depend on and shows them in one view: on the web, as a Windows app and as an Android app. The two installed apps also alert on a personal watchlist.

ROLE: Product, Design & Engineering
TIMELINE: 2 Weeks
YEAR: 2026
TEAM: Solo

(MY ROLE)

Designed and built Pulse alone: the React dashboard, readers for 7 feed formats, the landing page and both installed apps.
Cut Android's wait for a new-outage alert from up to 5.5 min to about 20 s while the phone is awake, for 72 of 77 feeds.
Stopped one Cloudflare incident re-alerting on Windows: 19 identical toasts in 4 hours before the fix, 1 alert and 17 quiet minutes after.
Wrote the release gate every version has needed since v1.0.16: it builds and tests all 3 surfaces, then launches both installed apps.

(WHY IT EXISTS): A deploy fails, and the first question is whether it is your code or a provider. The answer is spread over a dozen status pages. Pulse reads those pages for you.
(A FEED THAT FAILS IS NOT GREEN): Pulse shows only what each provider publishes and never tests a service itself. A feed that fails, times out or changes shape reads as unavailable. A reading older than 5 min leaves the counts, and no service gets a city it did not name.
(28 SERVICES BECAME 77): Pulse launched with 28 services. On September 21 the catalogue grew to 77, and every new feed was fetched before it was added. Only 27 of the 77 then showed on the map, since status pages name parts "API" or "Webhooks", not places. A service with no region now goes to a Worldwide hub.

Process

01
Three Surfaces

Day one shipped one React bundle as a Vercel site, an Electron app with its own local server and a Capacitor APK. 31 releases went out in 14 days.
02
One Design System

The phone layout was the desktop, shrunk: text down to 6 px and 41 controls under 44 px. One token file took over colour, type and spacing. An audit of 9 screens then found neither.
03
Scale & Latency

A full pass over 77 feeds cost 3.4 MB and 29.9 s, so an outage alert could wait 5.5 min. On an awake phone, Android now reads the 262-byte Statuspage indicator every 20 s, and a whole feed only when it moved.
04
Parity & Trust

Only Android said "back to normal"; Windows never did. One table of 36 cases now runs in both the Node and the Java tests. Quiet hours, a Quick Settings tile and a transfer code followed within a day.

How It Works

All three surfaces run the same monitor over each provider's own status feed. The two installed apps also decide when to alert, with one shared set of rules and one table of test cases.

The Dashboard

The same dashboard runs on all three surfaces: counts, a live incident inbox, a world map of 22 hubs and a directory of every service. Colour is reserved for status. Both screens are the live app, reading the feeds as they stood on September 30, 2026 at 06:04 UTC.

The Pulse Overview in light mode, reading of September 30, 2026: 77 services tracked, 68 operational, 8 active disruptions and 6 on your radar, the incident map with 3 outage, 10 degraded and 3 maintenance chips over the globe, and Live incidents, 19.

The Global map in dark mode, same reading: 3 outage, 10 degraded and 3 maintenance chips, pins on named hubs, and the Service conditions panel, headed Provider signals, not inferred city outages.

In The Detail

Each service keeps 30 days of what this device saw, and the strip says what grey means. An insight sets the provider's own words beside what you might notice and what to check next.

Cloudflare's detail sheet: Degraded, Minor Service Outage, and the 30-day strip, ringed in red, over its caption: Recorded on 1 of the last 30 days, while Pulse was open on this device. Grey means Pulse was not watching, not that the service was healthy.

A Dependency insights card for Cloudflare, Degraded: What's happening, Cloudflare One Clients are incorrectly challenged on some sites; What you might notice; What to check next.

(SYSTEM · ALERTS)

The Installed Apps

An alert names the service and quotes the provider's first open incident: a toast on Windows, a high-importance notification on Android. Both are staged here with the September 30 reading, in the apps' own words.

(ONE INCIDENT, ONE ALERT): Cloudflare publishes a component per datacenter, and its maintenance windows rotate through them. The Windows app took each rotation as news: 19 identical alerts for one incident over 4 hours. Components now count by level, not by datacenter, and maintenance is left out. After the fix: 1 alert, then 17 quiet minutes where the old rule would have sent 5.

(SCOPE)

Android notification channels: 4
kinds of toast on Windows: 3
home-screen widgets, plus a tile: 2

Pulse on Android: the notification Cloudflare · service issue, Cloudflare One Clients are incorrectly challenged on some sites, the ongoing Pulse is monitoring your watchlist, and the Pulse / My stack widget, 2 watched services reported issues, OpenAI and Cloudflare Degraded first.

Pulse on Windows, with another app open: the toast Cloudflare · service issue with the same incident, above the tray menu Open Pulse, Watchlist, Quit Pulse and the Pulse tray icon.

WHERE EACH SURFACE CHECKS

Website: Reads /api/status every 30 s while the tab is visible. Sends no alerts.
Windows: A tray monitor checks the watchlist every 30 s after the window closes.
Android: Every 20 s while the phone is awake, a small status read for 72 of 77 feeds. Otherwise an alarm, every 5 min at best.
Widgets only: WorkManager redraws them every 15 min, as often as Android allows.

The Website

The landing page at pulses4u.in walks through the app, 8 moments when a developer needs an answer, what the installed apps add and 7 questions. Every screenshot on it is a capture of the live app.