Product · UX · 3D · Detection logic
A one to one model of the farm, and the panel a grower runs it from
3 min read
Role
Product, design and build
Timeline
2026
Scope
Product · UX · 3D · Detection logic
Industries
Indoor vertical farming
Outcome
Live on a working farm as the control panel for environment and plant health
The grow room as a 3D model, each shelf carrying its own live reading
The panel a grower runs the farm from
A digital twin is a live one to one model of a real place, accurate enough that what is on screen is what is happening in the building. AGEYE’s twin is the grow room in 3D, with live environment readings and plant health sitting on the shelf they belong to. It is live on a working farm, and it is the panel a grower opens to run it. I specced, designed and built it over three months, hands on, working with AI coding agents.
A walk through the twin, from the whole room to a single shelf
The brief
Indoor farming has spent a decade getting off clipboards and onto screens. The data arrived and then it stopped, as rows in a table. A grow room can hold more than sixty thousand plants on a fixed schedule, and the conditions that decide whether they make weight, air temperature, humidity and CO2, have no appearance. A grower can stand in the middle of that room and see none of it.
The racks are identical. The air around them is not. Air conditioning, humidity and airflow leave the top shelf warmer than the bottom, air standing in one aisle, a cold pocket nobody predicted. A plant grows in the conditions at its own position, not the building’s average. A table can say that something is wrong somewhere. It cannot say which of two hundred shelves to walk to.
AGEYE had wanted this for three years. The farm map shipped in 2024 and got as far as rack level, built from pre-rendered 3D. Past that it stalled: nobody had the time, the data model was not there, and nothing could diagnose what a reading meant.
The same data as a table and as a place
Plant health on the same twin: every plant on the rack, measured on every scan
The solution
The 3D room is generated rather than modelled by hand. It is produced by the farm builder and arrives carrying its own configuration, so anyone working on the twin can create a farm and load it without a designer in the flow. On that model I built a visualisation for each class of data the farm produces: heat shelves carrying temperature and humidity at the level they were recorded, tank readings shown in position, and navigation that resolves from the whole room to a rack, a level, and an individual position.
The second layer determines whether any of it is acted on. An alert reading “air temp 29.4, above 28” reports a symptom and leaves the diagnosis to the grower. Attention resolves four questions instead: what, where, how large, and whether the cause is the room or the sensor. It reaches that verdict through three independent tests, against a wired reference probe, against the same position at the same hour on previous days, which prevents the daily lighting cycle being reported as a fault, and against the sensor’s own recent behaviour for drift and battery decline. Where the evidence does not support a verdict, the finding returns “cannot confirm” and names what is missing, because an alerting system that rounds a guess up to a claim is abandoned within a month.
Plant health was then brought onto the same model. Plant alerts surface in the twin and link back to the plant health record, so crop condition and environment are read in one place rather than two. That is what makes this a control panel rather than a view: everything the farm records, presented on the room in which it is happening.
From the room to a rack, a level, and one position
One finding per row, in the words a grower would use
to build, after three years of attempts that never landed
named microclimate patterns the system recognises
verdicts on every finding, including cannot confirm
The impact
Acting on where, not just what
A grower can name what the air is doing at a single position, in a room of sixty thousand plants, in an environment that changes hour to hour. Pinpointing a microclimate that small was not possible on this farm before.
What it takes out is the search. Finding a problem used to mean walking the room, checking screens and forming a judgement from numbers that never named a place. Now the place is named and the cause is already diagnosed as the room or the sensor, so whoever goes to fix it knows which before they leave. That is hours off an investigation, and it shortens the gap between a microclimate forming and someone correcting it, which is where crop loss and short shipments start.
A grower no longer has to hold the building in their head.
One shelf named, the finding behind it, and the way there