Jackson Ranucci

RS-1 Dashboard

Turns a surprise breakdown into scheduled work.

Year
2026
Status
Simulated data
Built with
  • TypeScript
  • React
  • Vite
  • Tailwind CSS
  • Supabase
  • MQTT
  • Vercel Functions

The problem

Equipment failure is not the expensive part. Being surprised by it is.

Mechanical equipment fails constantly, and no amount of maintenance stops that. What makes a failure costly is having no warning it was coming: the work becomes an emergency, it happens on the equipment's schedule rather than yours, and by the time anyone looks, a small problem has usually turned into a large one. The failure was never really avoidable. The surprise was.

What it makes possible

  • 01

    Time to fix it before it becomes major

    Degradation is caught while the equipment is still running, so the repair happens as planned work at its current size rather than as an emergency after it has grown.

  • 02

    The cause and a fix, not just an alarm

    Each warning names the channel driving it and carries a recommended remedy, so the work can be scoped and the right parts brought on the first trip.

  • 03

    A grade on how much to trust it

    Every prediction is scored on the strength of the evidence, how wide the window is, and whether past warnings on that same asset were borne out. The verdict is advice, not a number: plan on it, confirm on site first, or do not act on this alone.

  • 04

    It tells you when to stop trusting it

    A frozen sensor, channels contradicting each other, or a poor track record on that machine will say so and recommend a qualified eye in person. Data problems override a high score outright — a confident number built on a dead sensor is worse than an uncertain one.

  • 24 h+Warning ahead of a simulated failure
  • 12Assets monitored
  • 2 monthsRunning continuously
The RS-1 fleet dashboard listing twelve monitored assets across four sites, each row showing status, health percentage, vibration and temperature, beside a fleet health trend and a recent alerts panel.

In use

The Confidence page scoring an asset 95 of 100, with a plain-language verdict, a breakdown of how the score is built, and a live per-channel rate of change.
Confidence is its own page. It scores how much to trust a prediction and shows the reasoning, rather than asking anyone to take the number on faith.
The asset drawer opened over the dashboard for Pump A, showing per-channel diagnostics, a health gauge, failure risk, a predicted failure window, and trend charts for vibration, temperature and current draw.
Clicking any asset opens its diagnostics in place — channel status, health, failure risk and the predicted window, without losing the fleet view.

Where it stands. The dashboard is deployed and has been running continuously for a couple of months against a simulated plant, where it has repeatedly called a failure more than a day before it happened. It has no live customers yet: the HTTPS ingestion endpoint is built and waiting for its first real site to point sensors at it.

The part worth pointing at. Most predictive tools hand back a date and let you work out whether to believe it. This one grades its own certainty and says what to do about it — and when the evidence is thin, or a sensor looks frozen, or its own past calls on that machine were wrong, it says so and recommends having a professional confirm before anyone commits downtime or parts. A prediction that admits its limits is worth more than one that never does, because the crew keeps listening to it.

The engineering underneath that was making sure only one answer exists. The dashboard, the alert, the email and the push notification all run the same predictor and the same confidence scorer, so the drawer can never read “stable” while an alert claims the machine is about to fail.