Deep Technical Architecture

Physics-Informed AI
for Building Energy Optimization

Unlike black-box ML, our AI understands the physics of your building. It solves thermodynamic equations — never hallucinates, always respects the laws of physics.

System Architecture

How It All Works Together

Core Technology

Physics-Informed
Neural ODE

Our core is a Neural Ordinary Differential Equation (Neural ODE) — a deep learning architecture that solves continuous-time differential equations. Unlike standard neural networks that learn correlations, our model learns the actual thermodynamic behaviour of your building.

Pre-Trained on Simulations

The model arrives 95% ready — trained on thousands of EnergyPlus & OpenModelica simulations before touching your real system.

Constrained by Physics

Every prediction respects conservation of energy, heat transfer laws, and thermodynamic constraints. No hallucination possible.

1–3 Day Calibration

Shadow Mode fine-tunes the model to your specific building. Zero risk — proves value before going live. No months of BMS data required.

Model Predictive Control

The Optimizer That Thinks
12 Hours Ahead

The Digital Twin feeds into our MPC optimizer — a mathematical framework that computes the optimal HVAC schedule every 15 minutes, looking 12 hours into the future.

WEATHER

12-Hour Forecast

Integrates GFS/ECMWF weather models with per-zone solar gain calculations by orientation (N/S/E/W).

TARIFFS

Dynamic Pricing

Shifts thermal loads to off-peak hours. Pre-charges building’s thermal mass during cheap periods — uses it as a free battery.

COMFORT

Occupant-Aware

Maintains strict comfort bounds ±0.5°C MAE while minimizing energy cost. Never sacrifices comfort for savings.

Infrastructure

Edge-First
Architecture

All computation runs on the local edge gateway. Your building data never leaves the site — full EU data sovereignty compliance.

On-premises deployment — No cloud dependency. Works offline. Continues optimization even without internet connectivity.

Data sovereignty — Compliant with EU data protection requirements. Building sensor data never leaves your network perimeter.

Protocol-agnostic — BACnet, Modbus RTU/TCP, KNX via Virtual BMS gateway. Connects to your existing infrastructure.

Hardware-agnostic — Connects to any boiler, chiller, valve or fan. No HVAC replacement required.

See the AI in action
with your building

Book a free 30-minute demo. We’ll walk through how our Neural ODE digital twin models your building’s thermodynamics and calculate your estimated savings.