Science
Methodology, cited end-to-end.
We run an explicit model bench. Every prediction the platform makes is traceable to the model, version, dataset, and confidence interval that produced it. Where research models exist, we use them. Where domain physics exists, we use that too. We do not train novel foundation models — we orchestrate the best of what the field has produced.
Model stack
Open foundations, cited and named.
We orchestrate the best open and research-grade models for energy — never black boxes. Every prediction, every dispatch decision is traceable to the model, dataset, and confidence interval that produced it.
Read the methodologyTimesFM 2.5
Google Research
Load + generation forecasting
Day-ahead and seasonal forecasts of electricity load, solar/wind generation, EV charging, heating/cooling demand. Zero-shot on new sites.
Chronos
Amazon
Probabilistic time-series ensemble
Quantile forecasts of household and feeder load. Used in ensemble with TimesFM for confidence bands.
Moirai
Salesforce
Universal forecasting (encoder)
Mixed-frequency forecasting across any-to-any horizon. Strong on irregular SCADA telemetry.
Temporal Fusion Transformer
Open research
Interpretable multi-horizon forecasting
Sub-hourly substation load forecasts with feature attribution — the workhorse for explainable demand projections at the feeder level.
GraphCast
Google DeepMind
Medium-range weather prediction
Drives load and renewable generation forecasts: storms, heat waves, cold snaps, wind speed at hub height, irradiance at PV plants.
Qwen3-VL
Alibaba
Vision-language reasoning
Drone and pole-camera inspection: cracked insulators, vegetation encroachment, corroded hardware, thermal hotspots. Reads field images and writes structured findings.
Pixtral
Mistral
Lightweight VLM for edge devices
On-device inspection from substation cameras and lineworker phones. Fits a Hailo-10H or Jetson Orin Nano Super at line-power budgets.
SAM 3.1
Meta AI
Spatial segmentation
Vegetation masks for ROW management, asset bounding boxes for inventory, footprint extraction for solar siting.
PANNs
Open research
Audio-event classification (transformer condition)
Acoustic anomaly detection on transformers and switchgear: partial discharge, mechanical faults, corona, bearing wear. Continuous, low-power, in-pole.
YAMNet
Edge audio embeddings
Always-on acoustic monitoring at the cabinet edge — emits embeddings for downstream classifiers without sending raw audio off-site.
GridSim ensemble
Ipnops · open foundations
Power-flow + DER orchestration simulator
What-if simulation: VPP dispatch, FLISR re-routing, microgrid islanding, BESS sizing, EV smart-charging, demand-response events. OpenDSS / GridLAB-D family.
Gemini 3.1 Flash Lite
Google DeepMind
Reasoning + tool orchestration
Conversational interface, tool-calling agent for forecast/simulate/dispatch, document QA over standards (IEEE 2030.5, IEC 61850, NERC CIP), report generation.
Gemini Embedding
Google DeepMind
Semantic search over operational text
Standards corpus, manuals, work orders, outage post-mortems. Matryoshka 1536-d for pgvector halfvec at the workspace tier.
Benchmarks
Tested against the published baselines.
Forecasting models are evaluated zero-shot on a fresh site before deployment, then fine-tuned only when the zero-shot MASE exceeds the seasonal-naïve baseline. On the GEF-Com / BDG2-style household-load benchmarks we track in 2026, top foundation models are now reaching MASE values near 0.31 at long context lengths — a ~47% reduction over seasonal naïve.
Inspection vision is evaluated on a customer-curated holdout of substation, line, and ROW imagery. We deploy Qwen3-VL or Pixtral depending on the silicon at the edge: Pixtral fits within the 25 W envelope of a Hailo-10H card; Qwen3-VL runs comfortably on a Jetson Orin Nano Super.
Acoustic anomaly detection (transformer partial discharge, switchgear corona, bearing wear) uses PANNs CNN14 embeddings with a per-asset linear classifier — the platform fingerprints each transformer once, then watches its acoustic signature continuously.
Standards
Native to the protocols of the grid.
The grid runs on a thicket of standards. A platform that doesn’t speak them natively is doomed to perpetual integration work. We speak them natively.
IEC 61850
Substation automation, GOOSE, Sampled Values
IEC 61968 / 61970 (CIM)
Common Information Model — CGMES profiles
IEEE 2030.5 + CSIP
Smart inverter / DER comms with TLS + PKI
IEEE 2800-2022
Inverter-based-resource ride-through requirements
IEEE C37.118
Synchrophasor (PMU) wide-area monitoring
OpenADR 3.0
Automated demand response over JSON / REST
OCPP 2.1 + ISO 15118-20
EV charging, V2G, Plug & Charge
OpenFMB / DNP3
Field-message bus, distribution telemetry
Matter / Thread
Residential interoperability for energy devices
NERC CIP-003-9 (2026)
Cyber-system governance for low-impact BES
FERC Order 2222
DER aggregation in ISO/RTO markets
ISO 50001 / ISO 50006
Energy management for buildings & portfolios
Selected references
Where the work comes from.
Decoder-only Foundation Model for Time-Series Forecasting
Das et al. — TimesFM, 2024 (rev. 2026)
Chronos: Learning the Language of Time Series
Ansari et al. — Amazon, 2024
Unified Training of Universal Time Series Forecasting Transformers
Liu et al. — Moirai, 2024
Temporal Fusion Transformers for Interpretable Multi-horizon Forecasting
Lim et al., Int. J. Forecasting 2021
Learning skillful medium-range global weather forecasting
Lam et al. — GraphCast, Science 2023
Qwen3-VL: A Multimodal Large Language Model Family
Qwen Team, Alibaba 2026
PANNs: Large-Scale Pretrained Audio Neural Networks
Kong et al., IEEE/ACM TASLP 2020
UNIFI Specifications for Grid-Forming IBRs
NREL, 2024
IEEE Std 2800-2022 Compliance: GFM Negative-Sequence Currents
Nilsson et al., IEEE Trans. PE 2024
Foundation Models for Short-Term Load Forecasting on Consumer Hardware
ACM e-Energy 2026