CAT Operator Assistant Dashboard
Product design | Dashboard design
2026
CAT Operator Assistant is a concept cab tablet for Caterpillar excavator operators and their fleet managers, built at a hackathon. It turns data the machine already records into five glanceable screens, so an operator can check pace, safety and machine health without leaving the seat.
Services:
Challenge:
An operator makes dozens of decisions an hour with almost nothing to go on.
Role:
Product design and front end, end to end: the interaction model, every screen, the procedural 3D excavator and the job-time estimator.
01
About
Most machine dashboards are built for the office and squeezed into the cab, so I started from the seat and worked outward.
I built this at a hackathon as a concept for Caterpillar: a tablet in the cab that turns data the excavator already records into five screens, plus a read-only view for the fleet manager. No new hardware on the machine.
It runs as a static web app in plain HTML, CSS and JavaScript so it loads on a weak site connection, and the excavator is generated in code with Three.js rather than downloaded as a model.
02
Details
- Type
- Conceptual project for Caterpillar
- Stack
- Vanilla HTML, CSS and JS, procedural Three.js (r128)
- 3D
- Excavator built in code, no downloaded assets
- Repo
- github.com/Aayushvz/cat-operator-assistant
Challenge
The day's jobs live in their head, the machine's real condition sits in gauges nobody reads until something breaks, and training happens when a supervisor finds the time. A new operator cannot tell if they are behind, working safely or wearing the machine out.
The fleet manager sees even less. A seatbelt violation or a job running two hours late reaches them after the shift, when the pattern behind it is gone.
Solution
I designed for the cab first and the office second. Each screen answers one question an operator asks all shift: am I on pace, is the machine safe, what is next. Status runs on red, amber and green, the most important number is always the largest thing on screen, and every target is sized for a gloved hand.
The fleet view reads the same data model, read-only, across every machine on site. And every number says where it came from: the real telemetry is used as given, and anything simulated is labelled sample data, never passed off as a live sensor.
03
Start of shift
I made the sign-in double as a pre-start safety check, so a shift cannot begin in an unsafe state.
Skill level sets the day's estimates
The operator picks a profile, beginner to expert, and confirms with a PIN or a badge. The level is not decoration: it sets the expected time for every job that day, so the estimates match the person in the seat.
The engine stays locked until the belt is on
A blocked start is logged as a safety event, and the belt is watched all shift. If it comes off during a wait, the screen warns the operator before the machine moves again, so the fleet owner gets a guarantee rather than a policy.
04
Home
Home is built to be read in a glance between loads.
A 3D machine instead of a sensor table
I drew the excavator live in 3D and coloured it by condition, red for a fault and amber for check soon, with six tappable parts. An operator reads a red boom faster than a fault code.
Pace is measured against the operator's own best time, not anyone else's.
05
My tasks
Plans change on site, so the job list replans itself.
A plan that replans itself
Now, Next and Shift-left sit above the list, with one-tap Mark done and Running late. The plan checks the forecast and moves wind-sensitive work like demolition to another day.
The job-time estimate is a range, shown beside how accurate past estimates were, rather than one confident number an operator learns to ignore.
06
Safety and reports
Safety is designed for the worst moment on site, including the one with no signal.
One hold for SOS, even offline
Holding SOS for 1.5 seconds stops the machine, alerts the supervisor and saves the last 60 seconds of machine data. An incident report takes a hold and two taps.
Proximity zones grow in rain, and every report saves to the tablet with a machine snapshot until signal returns.
07
Fleet view
The manager's view reads the same data the cab tablets send, read-only.
Who needs attention today
Managers filter by operator, machine and date, and a review list flags each operator's habit changes with one-tap Talk to operator or Dismiss. Habits are measured against that operator's own baseline, so the conversation is about their week, not a leaderboard.
08
Results
Measured on the same five tasks the estimator learned from, and the app says so on screen.
09
What it taught me
What I learned
My first drafts were a dashboard, not a cab.
They looked like a generic AI dashboard with tinted cards, so I threw them out. A gloved hand in a cab needs fewer, bigger and louder things, and every screen had to pass that test before it looked finished.
An estimate is only useful if it is honest about itself.
The estimator's 2.4% error was measured on the same five tasks it learned from. I put that on screen instead of hiding it, because an operator who catches a model overselling itself stops trusting the rest of the tablet too.
Keep exploring
