Orbit, ML monitoring on mobile. App design for ML teams.
Rajaraman Arumugam
Orbit mobile. Full model health data in your pocket, not a push notification summary.
The mobile app surfaces the actual numbers: p99 latency, accuracy delta, drift score, inference volume. Every deployed model, live status, same data as the desktop dashboard.
Four screens: fleet-level model health overview, per-model detail view with accuracy timeline and drift tracking, inference log with anomaly filtering, and per-model alert threshold configuration.
Inside the app
The Models Overview loads all seven at once. Six green. One red. Fraud-detector-v3 sits at the top, status badge ALERTING, accuracy at 91.2%, drift delta shown inline. Full fleet at a glance, no scrolling required.
Tapping a model opens the detail screen. 90-day accuracy timeline, current drift score, p99 latency, and all active alerts with timestamps. The same data panels from the desktop view, adapted for one hand. Nothing cut.
The Inference Log shows the last 200 calls in order: timestamp, model version, latency per call, confidence score, anomaly flag. Each anomaly is flagged inline — no log parsing, no separate query.
Settings handles per-model alert configuration: SLA thresholds, accuracy drift alerts, anomaly sensitivity per model. Workspace defaults apply unless overridden per model.
The Inference Log is the screen that earns the app. Not the overview, not the alert badge. Most monitoring tools send a push notification and leave you to find the context yourself. Orbit surfaces the exact call that crossed the threshold. That's the difference between a 3-minute investigation and a 45-minute one.
Does your team have mobile access to model health when something alerts outside business hours?
Raw UX
We design products, interfaces, and design systems for startups and SaaS teams. Working on a new product? Let's craft the design together.
Like this project
Posted Sep 26, 2026
Orbit mobile. Full model health data in your pocket, not a push notification summary. The mobile app surfaces the actual numbers: p99 latency, accuracy delta, drift score, inference volume. Every deployed model, live status, same data as the desktop dashboard. Four screens: fleet-level model health overview, per-model detail view with accuracy timeline and drift tracking, inference log with anomaly filtering, and per-model alert threshold configuration.
