Meter Data to Decisions: Grounded LLMs for Net-Zero Operations
Project ID: 2531bd1609
(You will need this ID for your application)
Research Theme: Energy and Decarbonisation
Research Area(s):
Artificial Intelligence Technologies
Built Environment
Digital Twins
UCL Lead department: Bartlett School of Environment, Energy and Resources (BSEER)
Lead Supervisor: Dimitrios Rovas
Project Summary:
Why this research is important
Buildings brim with data, yet teams still struggle to share the correct facts at the right moment. With LLMs, we can finally bridge disciplines — if we provide rich context and overcome poor interoperability. We also need inclusive interfaces that provide people with the information they need, in the format they require, when they require it. Your work will help develop tools that can have a real-world impact; you will be able to use and validate these tools to support UCL’s efforts to decarbonise its campus by turning meter data into targeted, trustworthy actions.
Who you will be working with
You’ll work with academics from the Bartlett and Chemistry, and collaborate with Estates and Facilities. You will have unparalleled access to campus thermal and electrical metering data, as well as related asset/process information. You will have access to very significant computing resources.
What you will be doing
- Design a context engine that captures processes, roles, and information needs in knowledge graphs.
- Build neuro-symbolic AI that blends ontologies with machine learning so LLMs give specific, non-generic guidance.
- Develop graph-RAG pipelines that ground answers in building data and standards.
- Create interfaces (FM co-pilot, APIs/dashboards) that diagnose user/system inefficiencies and recommend actions for monitoring and facility management.
- Quantify benefits for carbon, comfort, cost, and decision quality, contributing credible evidence for the Infrastructure Master Plan.
Who we are looking for
A curious, hands-on PhD student with a quantitative background in computing, data/AI, or building/energy modelling.
Essential: Python, data wrangling, ML fundamentals, and interest in LLMs and knowledge graphs, curiosity-driven learning.
Nice to have: prior knowledge of working with built environment data, use of energy modelling tools like EnergyPlus/Modelica. You value reproducibility, responsible AI, and teamwork with real users.