The Engine

The real-time grid intelligence engine

Voltpathio's platform ingests ISO grid stress signals, BMS telemetry, and weather data to generate load-shaping recommendations at 200ms decision cycles.

Request Access See How It Works
Voltpathio platform architecture: data ingestion layers flowing into ML forecast engine and dispatch layer
Architecture

Data ingestion → prediction → dispatch

1 ISO/RTO Real-Time LMP Feed (PJM, MISO, CAISO, ERCOT, SPP, ISO-NE — 1s resolution)
2 Weather & Solar Irradiance Forecast (NWS HRRR model + site GPS irradiance)
3 BMS Telemetry Adapter (BACnet/IP, Modbus TCP, Project Haystack)
4 Demand Forecast Engine (gradient-boosted ensemble, 60s inference cycle)
5 Comfort Constraint Solver (hard temp limits, occupancy schedule, tenant SLA rules)
6 Load Command Dispatcher (200ms end-to-end, BMS write via BACnet WriteProperty)
Model Performance

94% peak demand forecast accuracy at 15-minute horizon

The demand forecast engine is a gradient-boosted ensemble trained on 18 months of ISO grid event data across PJM and MISO, cross-referenced with building thermal mass coefficients, occupancy schedules, and National Weather Service mesoscale analysis data. It outputs a probabilistic peak demand estimate with confidence intervals every 60 seconds.

Accuracy degrades to 85% at 45-minute horizon — still sufficient to issue HVAC pre-conditioning commands before the demand window opens. We do not claim accuracy we haven't measured: the 30-day pilot generates site-specific backtesting against your actual billing intervals so you can verify the numbers before committing to a full subscription.

60% 75% 88% 94% 98% 15 min 92% 30 min 85% 45 min Forecast accuracy by time horizon
Infrastructure Grade

Built for operational infrastructure

Encrypted at rest and in transit

AES-256 data encryption at rest. TLS 1.3 for all API and BMS adapter traffic. Designed with SOC 2 Type II controls — audit report available to Enterprise customers on request.

Granular role-based access

Per-building permission scopes with read-only, operator, and admin roles. Immutable audit log for every load command issued — timestamp, building ID, zone, kW delta, and operator identity recorded. Operators can override any automated command in real time.

99.95% uptime SLA on command dispatch

Multi-region cloud architecture with active-active redundancy on the dispatch layer. No single point of command failure. If the cloud layer drops, BMS adapters fall back to pre-configured local guardrail rules — demand protection continues even during connectivity loss.

Get Started

Ready to connect your first building?

BMS connection in under 48 hours. See your demand charge exposure in the first session.