Six sectors.
One closed loop.

The measurement is settled; the integration is the problem. BACnet holds roughly 77% of the global BMS market1 and still doesn't get you interoperability, because point names are whatever the commissioning contractor typed in 2011.

We buildBACnet/IP and Modbus gateways that speak to Siemens, Honeywell, Schneider and Distech panels; point-mapping and normalisation; fault detection rules plus learned baselines; submetering hardware where the existing plant has no instrumentation.

Insurer data puts the US fleet 8.6% below P50 energy estimates2, and soiling alone runs 0–15% depending on site3. Meanwhile 108 GW of battery storage went in worldwide in 2025, eleven times the 2021 installed base4 — all of it needing monitoring nobody has built yet.

We buildSunspec and Modbus inverter polling, string-level current monitoring boards, soiling and shading detection from production data rather than extra sensors, BMS-to-cloud gateways for battery racks, and alarm logic that distinguishes a cloud from a blown fuse.

The pattern is consistent elsewhere: Gartner finds only 48% of AI projects reach production, taking a median eight months to get there5. Almost none of that gap is model accuracy. It is PLC integration, line takt time, lighting, and who gets paged at 3am.

We buildIn-line inspection cells with controlled lighting and telecentric optics, PLC handshakes over Modbus/TCP and Ethernet/IP, edge inference on Jetson or industrial x86, reject-arm timing, and an HMI a line operator can override without a laptop.

Trucking operating cost hit a record $2.336 per mile in 2025, with repair and maintenance up 8.6% year on year6. Meanwhile 79% of new cars sold worldwide in 2024 shipped with embedded telematics7 — which mostly means the OEM has your data and you don't.

We buildOBD-II and J1939 dongles with LTE-M or NB-IoT backhaul, CAN decoding for mixed fleets, driver-behaviour scoring that survives a rough road surface, geofencing, and predictive-maintenance features built on DTC history rather than mileage alone.

The same paper notes CEA supplies under 1% of US food crops while consuming more energy than all open-field cultivation combined. That is the whole business case in one sentence: control the climate precisely or the economics don't close. Greenhouse lettuce uses roughly 21 litres of water per kilo against 201 in a California field8.

We buildMulti-zone climate controllers with PID loops that don't fight each other, EC/pH dosing with redundant sensing, VPD-driven venting and screening, DLI-aware supplemental lighting schedules, and a fallback mode that keeps the crop alive when the network is down.

AMRs, inspection crawlers, agricultural platforms and gantries. The discipline transfers directly — a greenhouse controller and a mobile robot are both closed-loop systems that have to fail safe — which is why we don't treat robotics as a separate business.

We buildLocalisation and mapping, motion planning under real payload, sensor fusion across LiDAR, radar, wheel odometry and IMU, control tuning against the machine rather than the simulator, and a stop chain a supervisor can reach without training.