Forecasting models
Time-series and regression models for demand, yield, consumption and other quantities that change over time.
Services
We build machine learning systems that use historical and real-time data to estimate what happens next, catch unusual patterns and support better decisions. Where the problem needs it, we add graph and spatial context, so models know how things are connected and where they are. Every project is tested against how the work is done today.
Applied in client work for Asset Circles and in our FlyingVink and Vink Farming pilots. See Asset Circles case →
Time-series and regression models for demand, yield, consumption and other quantities that change over time.
Baselines per asset, building or site, and detection of deviations that deserve attention.
Models that label, score or rank, evaluated against outcomes you already know.
Extracting information from images, such as detecting and measuring objects.
Choosing between options under constraints, using the predictions as input.
Baselines per building and detection of unusual consumption and comfort patterns, as in our work for Asset Circles.
Forecast demand, yield or consumption and compare the result with your current planning method, as in Vink Farming.
Rank candidates by score, such as the buildings where drone cleaning is economically attractive, as in FlyingVink.
Detect and measure objects in images at a scale that manual inspection cannot reach.
Add graph features to a model, as ReviewGraph does to predict review ratings.
Map the data, define the problem in measurable terms and check whether the data can support an answer.
Develop a model or pipeline on your data and make its results visible.
Compare results with known outcomes and with your current way of working, then decide to deploy, refine or stop.
Integrate with your systems, monitor the model and retrain as the data changes.
Every project follows the same four steps. Understand → Build → Validate → Deploy. About our way of working →
Client work
Sensor, weather and equipment data turned into building-specific anomaly detection and advice.
Read the Asset Circles case →Pilot
Earlier signals on crop development and yield from imagery, parcel and weather data.
See Vink Farming ↗Pilot
Estimating where drone cleaning is economically attractive for a building.
See FlyingVink ↗The methods are standard. The difference is context: when a problem depends on how things are connected or where they are, we add knowledge graph and spatial methods, and we test every model against how the work is done today rather than only against a benchmark.
We first check whether your data can support an answer, before building anything. If it cannot, we say so and describe what data would be needed.
By comparing it with a baseline, your current way of working, and with outcomes you already know. The result is one of three decisions: deploy, refine or stop.
Both. A project can end at a validated prototype or continue to production pipelines, APIs, dashboards, monitoring and handover documentation.