Research direction

AI for Energy

Domain-specific AI for energy workflows, including LLMs and VLMs for causal, mathematical, and physics-grounded reasoning.

EnergyAI Domain LLMs Physics-grounded reasoning Subsurface workflows

Focus

This direction builds AI systems for exploration, production, and subsurface workflows where the model must reason with text, images, physics, uncertainty, and operational constraints.

Typical Questions

  • How can domain LLMs and VLMs reason with physical and mathematical constraints?
  • How can AI support causal and uncertainty-aware decisions in energy workflows?
  • How should multimodal models combine reports, diagrams, sensor data, and expert knowledge?
  • How can we evaluate reliability in specialized, high-stakes technical domains?
Selected papers

AI for Energy Highlights

3 papers
Large language model generating household routines and weather context distilled into energy load profiles
2025 · Household Energy Modeling

Knowledge Distillation from Large Language Models for Household Energy Modeling

  • Uses LLMs to generate realistic and culturally sensitive household energy scenarios.
  • Models family structure, weather, daily routines, HVAC operation, and energy signatures.
  • Creates a cost-effective route to scenario-based energy optimization under data-access limits.
Paper PDF
Net-zero buildings with batteries sending privacy-preserving updates to a federated reinforcement learner
2025 · Federated RL / Smart Grids

Generalising Battery Control in Net-Zero Buildings via Personalised Federated RL

  • Studies optimal energy management for building-based microgrids.
  • Uses collaborative, privacy-preserving RL to reduce energy cost and carbon emissions.
  • Shows Federated TRPO can be competitive without extensive hyperparameter tuning.
Paper PDF
Generator, auditor, and validator LLM agents governed by constraints to create household energy scenarios
2026 · Governed LLM Agents

WattCouncil: Context-Aware Household Energy Scenario Generation With Governed LLMs

  • Uses governed LLM agents to generate, audit, and validate household energy scenarios.
  • Emphasizes explicit constraints and scenario realism for privacy-limited energy data.
  • Connects language-model reasoning with structured energy-system simulation.
Publication list