AI platform
AI-powered biological solutions for smarter agriculture.
The platform exists to answer one question per field: will a biological input change the outcome here, and if so which one, at what rate, in which window. Seven models, one training set, one agronomist accountable for the answer.
Capabilities
What the models do.
01
AI crop health analysis
Multispectral imagery and stand counts are scored against the trial archive to separate a nutrition problem from a structural one before any product is discussed.
02
AI soil monitoring and prediction
Soil test history, texture and organic matter are modelled forward so the colonisation window can be predicted rather than guessed at planting.
03
AI disease detection
Leaf-level image classification flags the common cereal and rape pathogens early, and separates disease pressure from abiotic stress that biology cannot fix.
04
AI-based irrigation recommendations
Soil moisture curves and forecast evapotranspiration set the application window, because a foliar biological applied into the wrong water status does nothing.
05
Crop yield prediction
A response range, not a single number. The model returns the expected treated and untreated outcome with the uncertainty attached, including the fields where the range crosses zero.
06
Weather and climate risk analysis
Stress-window probability across the season sets the timing for StomaGuard, and flags the seasons where the risk is too low to justify the pass.
07
Smart farming dashboard
One field view: model scores, application cards, trial strips and measured outcome, so the season can be audited rather than remembered.
How it is trained
The trial archive is the training set.
412 replicated strip trials across nine states, each with an untreated strip. The failures are in the training data too, which is why the models are willing to predict no response.
| Model | Trained on | Refreshed | Reported as |
|---|---|---|---|
| Crop health | Imagery from 412 trials | Each season | Constraint class with confidence |
| Soil and colonisation | qPCR counts, soil tests | Each season | Predicted colonisation window |
| Disease detection | Annotated leaf images | Monthly in season | Pathogen class, abiotic flag |
| Irrigation and water status | Probe series, forecast ET | Daily | Application window in days |
| Yield response | Treated vs untreated harvest data | Each season | Response range, may include zero |
| Climate risk | Gridded forecast, 30-year history | Weekly | Stress-window probability |
No model output is published as a guarantee. Where a prediction and the measured result disagree, the measured result enters the archive and the model is retrained against it.
Data and governance
Whose data it is.
Field data stays the grower’s. It is used for that farm’s recommendations, and only enters the shared training archive in anonymised form with written agreement.
Advisory, never automatic
No model can trigger an application or an order. Every recommendation is reviewed and signed by a field agronomist.
Explainable output
Each score is returned with the inputs that drove it, so an adviser can disagree with it on the evidence.
Held to the untreated strip
Model accuracy is reported against measured treated and untreated outcomes each season, not against its own confidence.
The dashboard is opened for growers on a programme by their agronomist. There is no public sign-up, and nothing is sold through this site. Enquiries: contact page.