Follow The Logic
Mechanistic models of living systems, built to be steered
Every grower, every farm manager, every researcher already runs a model in their head. Light goes up, the crop does this. Water gets saltier, yield does that, but the fruit gets sweeter. Feed harder now, pay for it in three weeks.
Our job is to write that model down, make it run, and check it against what actually happened.
Week 20
Every truss, every fruit, at one moment of the season — from cropsimulator.com
It starts with the model, not the data
A model here is a set of differential equations that says how the system changes over time: what a plant does with today’s light and water, and what that does to yield, cost and labour twelve weeks from now. It is not a curve fitted to last year’s numbers.
That distinction matters when you have to decide something. A fitted curve tells you what usually happened. A mechanistic model tells you what happens if you change the setpoint, because the setpoint is in the equations.
We carry the chain end to end. A model that stops at biology is a paper. A model that stops at cash is a spreadsheet. Ours runs biology to steering to outcome, where the outcome is money, labour hours, water and energy use, and usually also emissions and animal welfare, because those are constraints the grower is actually held to.
Small data, narrow data, and the parts nobody measures
The data we get is rarely the data a statistician would ask for. Forty hand observations over a season. Three sensors logging every minute, measuring three of the thirty things that matter. A gap where the interesting week was.
The usual advice is to go and collect more. Sometimes that is right. More often the answer is to put knowledge where the data isn’t, and then be explicit about which parameters the data can actually pin down and which ones it cannot.
A model in days, a simulator in a week
New model in a few days. Running in a browser, with scenarios and KPIs, in about a week. That is not a stunt, it is the result of a framework that has built roughly fifty models already and reuses the machinery every time.
Dashboards, live data links and digital twins come after that, together with you and whoever already holds your data.
Where we work
- Crops and greenhouses — source and sink balance, leaf stress, salinity and water reuse. Public demo: cropsimulator.com
- Fish and aquaculture — feeding, stress, growth, disease pressure, and what an early detection is worth
- Human physiology — same mathematics, exploratory for now
Prefer to see the machine before the sales pitch? The auto-* pipeline page describes how a model gets built, layered, reduced and served, and the tools page has small calculators you can run in this page, right now, with no login.