A modular architecture for biological production.
From crop recipe and water balance to QA, batch data and unit economics. The Core Unit links nine functional modules with standardised interfaces — organised below by discipline.
Nine modules, one system.
We have worked out a modular system architecture as the basis for engineering and validation. Select a module to see what it does and what it records. Design
Intake & input quality
Receiving, lot ID, supplier input, inspection and quarantine where needed.
Status: design; supplier and acceptance procedures defined per crop.
Standard connections
Where possible, a Core Unit gets standardised connections for power, water, drainage, network and data, racks, safety, instrumentation and QA. A module can then be replaced or duplicated without redesigning the whole software and process architecture.
Add capacity on evidence
Extra capacity is added when the previous module shows measurable utilisation, repeatability and contribution to margin — scale-up as a series of validated decisions instead of one large capital bet.
Every crop as a versioned recipe.
A recipe covers seed or propagation material, pre-conditioning, germination duration, climate envelope, light, CO₂, water chemistry, nutrient solution, irrigation rhythm, harvest window and QA release criteria.
Internal R&D
| Recipe domain | Variables we specify and measure | Primary KPI |
|---|---|---|
| Seed / propagation | Supplier, lot, moisture, germination %, sanitation route | Lot acceptance and uniformity |
| Germination | Time, temperature, oxygenation, water schedule | Dry-matter recovery and microbial control |
| Vegetative phase | Temperature, RH/VPD, PPFD/DLI, CO₂, spectrum | Growth and composition trajectory |
| Nutrition | Macro- and micro-ions, pH, EC, water chemistry | Uptake, composition and safety |
| Harvest | Length, timing, fresh mass, dry-matter % | Saleable dry matter and release quality |
| Stabilisation | Drying curve or another validated route | Water activity, dry matter, composition retention |
One water balance, four quality classes.
Water is treated as a controllable process input, not one undifferentiated “clean water”. The source and treatment route depend on local water quality, crop, specification, regulation and cost. Design
Balance: source + optional AWH + recycled water = crop uptake + evaporation + RO reject + cleaning + wastewater
Design measures that make inputs, water, air and access controllable.
Set-points are envelopes
Temperature, humidity, vapour pressure deficit (VPD), CO₂ where used, light intensity (PPFD), daily light integral (DLI), spectrum, photoperiod, airflow, dew point and dehumidification are crop- and phase-specific variables. Every optimum is a hypothesis until it has been measured reproducibly.
Barriers across every route
Water treatment, HEPA filtration, positive pressure, airlocks, cleaning, zoning and controlled access are process barriers along the routes seed, water, air, people, equipment, recirculation, biofilm and post-harvest. They reduce and control risk; they do not make a product sterile.
| System | What we measure | What we only claim with specific evidence |
|---|---|---|
| Water | Source quality, conductivity, flow, RO recovery and reject, pH/EC, lab results, use per kg released dry matter | Purity or universal water savings |
| Air & climate | Temperature, RH, dew point, differential pressure, airflow, filter status, energy | “Sterile” or pathogen-free |
| Biosecurity | Input lot control, zoning, procedures, relevant analyses, batch release criteria | Contaminant-free claims without method and scope |
| Energy | kWh, peak power, use per kg released saleable dry matter, meter and timestamp quality | “Energy-efficient” without a comparison basis |
Instrumented first. AI never holds safety authority.
A KPI cannot be optimised reliably if it is not measured with known uncertainty. Sensors are production assets, with calibration, quality checks, drift detection, timestamp control and replacement plans. Design
Safety PLC & hard interlocks
Emergency stop, doors, hard temperature and pressure limits, dry-run protection, critical permissives.
Always above AI; AI can never override safety functions.
Edge control
Sensor fusion, data quality, recipe execution, alarms, event logging and fallback.
Degrades safely and keeps running if the cloud or connection fails.
Bounded MPC
Model predictive control for climate and energy, optimising within limits.
Works only inside validated operating envelopes.
Cloud & fleet learning
Recipe comparison, model versions, anomalies, supplier and unit benchmarks.
Improves knowledge; holds no safety authority.
Autonomy ladder — every step has its own gate
Roadmap
AI control room & MPC
A planned optimisation layer for monitoring, prediction and process optimisation, above safe base control.
Roadmap
Robotic harvesting & handling
A future module and research line for tray handling and harvest — not part of the first pilot cell.
Concept visual
Larger modular sites
High racks, larger modular buildings and façade solar exist only as concept visuals and future scenarios.
Sensor architecture (examples): temperature, relative humidity, CO₂, differential pressure, airflow, pH, EC, flow, pressure, level, conductivity, light/PPFD, electrical power, water use, weight, dry-matter sampling and RGB (optionally multispectral) vision. Operational technology, edge, enterprise and cloud are separated, with logged access, backups and versioned recipes and models.
Wet harvest mass is not a commercial KPI.
Harvest timing is set by composition, dry matter, microbial status, handling losses and the customer’s requirement. Drying or other stabilisation is part of the recipe, because it changes water activity, nutrient retention and economics; any additive or treatment is checked for legal status in the target market.
Design
| Product route | Release criteria | Main risk | Economic KPI |
|---|---|---|---|
| Fresh feed | Dry-matter %, microbiology, visual, temperature/time | Rapid spoilage, logistics | € / kg released dry matter |
| Stabilised ingredient | Dry matter, water activity, core temperature history, microbiology | Overheating, nutrient loss | € / kg released dry matter after stabilisation |
| Fresh food (later) | Microbial, sensory and shelf-life parameters | Shelf life, cold-chain drift | € / kg released food |
Every batch ends with an explicit quality decision.
QA owns the release decision. Batch statuses: HOLDRELEASEREWORKREJECT. No release is based on an AI score or camera output alone; specification, sampling plan, laboratory method and an authorised QA decision belong to each product. Design
- Input
- Seed lot LOT-DEMO-A · supplier · certificate of analysis · incoming checks
- Recipe
- Recipe version R-DEMO-0.1 · set-points · nutrient recipe · control mode
- Process
- Timestamped sensor values · alarms · deviations · operator actions
- Mass balance
- Seed dry matter → harvest dry matter → rejects → stabilised → released output
- Utilities
- Water by class · kWh · peak demand · waste streams
- QA
- Sampling plan · lab method · result · LOD/LOQ where relevant
- Release
- HOLD → RELEASE after QA sign-off
- Customer
- Lot delivered · customer outcome · complaints or returns
Economics are part of the design.
€ / kg released saleable dry matter
For feed. For food: € per kg released saleable product, plus edible dry matter and nutrient units where relevant.
The physical flow first
Seed dry matter, losses, gross harvest, rejects and released output must reconcile before any financial model is trusted.
What we meter per batch
Seed or propagation material, energy, labour, QA and consumables, maintenance, rejects and waste handling.
Technology supplier, integrator or engineer?
Let’s talk about a specific module, an interface, a component or an engineering work package.