BEEBOP LABS

World models for energy

World models for energy

Grounded in physics

Grounded in physics

Tested against reality

Tested against reality

We build learned, physics-grounded models of energy systems, and study how to make millions of decisions a second, under uncertainty, on the real grid.

  • <5%

    RMSE Day-ahead

  • 98%

    Dispatch reliability

  • 3,000+

    CITATIONS

  • 200K+

    Assets contracted

The Problem

The grid is becoming a decision problem.

Energy is the most important problem of our time. Food, water, security and computation all depend on solving it. The flexible capacity to solve it is already installed at the edge of the grid. Using it means making millions of decisions a second, under uncertainty, across markets, within physical limits.

P10-P90 Forecast

Measured • Live

10 Jan

10 Feb

Today

Sequential decision making

Open challenge: scaling. the unsolved problem at the heart of the field. Harnessing ever-larger pools of assets while complexity grows super-linearly.

Multi asset and multi market optimisation

Open challenge: scaling. the unsolved problem at the heart of the field. Harnessing ever-larger pools of assets while complexity grows super-linearly.

Data Center energy management

Open challenge: scaling. the unsolved problem at the heart of the field. Harnessing ever-larger pools of assets while complexity grows super-linearly.

rESEARCH

Research Threads

Our research starts with problems from the field, pursued together with an international network of leading institutes and a working academic lab. These are living threads, not finished features. Some will reach the grid. All of them teach us how energy systems actually behave.

P10-P90 Forecast

Measured • Live

10 Jan

10 Feb

Today

T01

World models for energy

learned, physics-grounded models of how energy systems behave, from a household to a fleet to a market. They capture the underlying dynamics, not just patterns, so they generalise across hardware, geographies and market regimes. The same family of model-based planning that produced the frontier results in game-playing and autonomous systems, applied to the grid.

T02

Predictive asset detection.

trading and control where the world is stochastic and the model is not linear. Reinforcement learning, probabilistic forecasting and mathematical programming, combined with expert domain knowledge.

T03

Cross-asset and cross-market optimisation

co-optimising heterogeneous assets across multiple market products through a single compact representation. (Project ACCORD in the UK, EMUFlex in Belgium - note both are unconfirmed)

T04

Data-centre energy management

the fastest-growing load on the grid is changing how the grid operates. Data centres can be coupled with distributed capacity to offset local congestion, and they themselves are transitioning from grid burdens to assets.

T05

Sequential decision-making under uncertainty

trading and control where the world is stochastic and the model is not linear. Reinforcement learning, probabilistic forecasting and mathematical programming, combined with expert domain knowledge.

T06

Standardising energy software infrastructure

OpenConnect, Market Coupling, providing connectivity abstraction, Market Design (tariff designer), (Engie FlexHub project - tbd)

Research infrastructure

Frontier research needs standard rails. Alongside the threads above, we build shared infrastructure the field is missing: connectivity abstraction across device makers, interfaces that couple assets to markets, and tools for tariff and market design. Engineering in service of the science.

Selected work

Published, cited, in use

Our research starts with problems from the field, pursued together with an international network of leading institutes and a working academic lab. These are living threads, not finished features. Some will reach the grid. All of them teach us how energy systems actually behave.

The Bench

The people behind the models

This team has scaled flexibility platforms from zero to millions of assets, twice. Backgrounds span Tesla, Palantir, Centrica, Goldman Sachs and Bain.

10+

Phds in AI, energy systems and optimisation

3000+

Academic Citations

15+

Years Operating GW-Scale Flexibility

2

VPP Exits

#1

Ranked VPP Technology Globally by Guidehouse

4

Continents of GW-scale operations

Chief product officer

Bert Claessens

3,000+ citations in physics-informed AI for power systems. Fifteen years of published research, in production.

Title

Name Nameson

trading and control where the world is stochastic and the model is not linear. Reinforcement learning, probabilistic forecasting and mathematical programming, combined with expert domain knowledge.

Title

Name Nameson

OpenConnect, Market Coupling, providing connectivity abstraction, Market Design (tariff designer), (Engie FlexHub project - tbd)

An international research network anchored by a working academic lab. Industry-led problems, peer-reviewed answers.

The Bench

The people behind the models

If they read like your problems too, come and help us.

Join the team

We hire researchers and engineers who want their work on the grid, not just on arXiv. No portal, no cover letters. Tell us what you've built.

Research collaboration

For labs, institutes and industry teams working on adjacent problems.