Hey there! I am a PhD candidate at the Department of Computer Science & Technology, University of Cambridge. My research, funded by the Gates Cambridge Trust, focuses on developing generalisable reinforcement learning (RL) models for the decarbonisation of building operations. More specifically, I am developing models that can rapidly adapt to new buildings without extensive retraining, aiming to bridge the gap between simulation success and real-world deployment.

Previously, I have explored the design and operation of sustainable energy systems for residential homes, incorporating technologies such as solar PV, stationary batteries, and bidirectional EVs.

Currently, my research is centred on meta RL, in-context RL and model-based RL to optimise the operation of HVAC systems in commercial buildings. I am visiting David Rolnick's and Yoshua Bengio's labs at Mila - Quebec AI Institute from March to September 2025.

Scholarships & Awards

  1. Gates Cambridge Scholarship
    The most prestigious postgraduate scholarship at the University of Cambridge, funded by the Bill and Melinda Gates Foundation. (October 2023 - September 2027)
  2. Heidelberg Laureate
    The Heidelberg Laureate Forum brings together the recipients of the most prestigious awards in mathematics and computer science with 200 selected exceptional young researchers from all over the world. (2024)
  3. Abbe Grant
    The Abbe Grant is issued from the Carl-Zeiss-Stiftung, an academic partner of the Heidelberg Laureate Forum Foundation (HLFF). (2024)
  4. IMSI Long Program Scholar
    The Architecture of Green Energy Long Program brings together selected researchers from all over the world to conduct collaborative research on the intersection of computer science and the energy transition over a period of ten weeks. (June - August 2024)
  5. Lauréate du Concours Général
    The Concours Général is an academic competition held every year between students in French High schools all over the world. Students who show great ability in one field are selected by their headmaster to represent the school internationally. Winners of the competition become life-long members of the Association. (2018)
Cambridge logo

Education

  1. University of Cambridge, Jesus College
    PhD in Computer Science (October 2023 - Present)
    Developing generalisable models for building control that can rapidly adapt to new buildings without extensive retraining, aiming to bridge the gap between simulation success and real-world deployment. Working towards creating the first large-scale transferable Reinforcement Learning (RL) framework for HVAC optimisation across diverse building types and climate zones. My research interests include: RL for HVAC control, in-context RL, decision transformers, meta learning, world models.
  2. ETH Zürich
    BSc Computer Science (2019 - 2023) Focus: Machine Learning, Computer Systems. Thesis: ”Designing PV-EV integrated Residential Microgrids in the post-COVID World”.
SAP logo Capgemini logo MAN ES logo

Working Experience

  1. Mila - Quebec AI Institute
    Montreal, Canada
    Visiting Researcher (March 2025 - September 2025)
    Supervised by Prof. David Rolnick, Prof. Yoshua Bengio and Dr. Donna Vakalis, working on model-based RL leveraging latent spatio-temporal world models for HVAC control.
  2. University of Cambridge
    Cambridge, United Kingdom
    Supervisions Databases - Department of Computer Science & Technology (September 2023 - Februar 2024)
    Teaching student supervisions for the Databases course taught by Dr David Greaves.
  3. Founderful
    Zürich, Switzerland
    Partner (September 2021 - October 2022)
    Part of the founding partners of the Campusfund, investing in student founders from ETH, EPFL and HSG.
  4. Archlet
    Zürich, Switzerland
    Software Engineer (June 2021 - September 2022)
    Working in the Product Team on the automation of end-to-end tests for the Archlet platform. Oversight of the software release processes.
  5. ETH Zürich
    Zürich, Switzerland
    Teaching Assistant for the Computer Science Network of Women (CSNOW) (2021)
    Teaching an introductory programming course to female High school students.
  6. ETH Zürich
    Zürich, Switzerland
    Teaching Assistant for Digital Design and Computer Architecture (Feb 2021 - July 2021)
    Teaching labs for the course Digital Design and Computer Architecture run by the SAFARI Research Group and led by Professor Onur Mutlu. I taught around 100 students on a weekly basis, correcting lab reports and supervising labs. I have also prepared and taught the exam preparation course for the entire student cohort.
  7. Ghelia - Sony Computer Science Laboratories
    Tokyo, Japan
    Machine Learning Software Development Intern (March 2019 - June 2019)
    R&D Section of the Professional System Solutions Division. I created a prototype for an adjusted WaveGAN implementation for synthesizing raw audio using generative adversarial networks. This prototype was later integrated into a larger AI-based solution used by Ghelia´s customers.

Publications

  1. HVAC-GRACE: Transferable Building Control via Graph Recurrent Neural Networks A. Berkes, D. Vakalis, D. Rolnick, Y. Bengio.
    38th Conference on Neural Information Processing Systems (NeurIPS 2024) - Tackling Climate Change with Machine Learning Workshop & Women in Machine Learning (WiML) Workshop
  2. HVAC-DPT: A Decision Pretrained Transformer for HVAC Control A. Berkes
    38th Conference on Neural Information Processing Systems (NeurIPS 2024) - Tackling Climate Change with Machine Learning Workshop & Women in Machine Learning (WiML) Workshop
  3. SOPEVS: Sizing and Operation of PV-EV-Integrated Modern Homes A. Berkes, S. Keshav
    ACM E-Energy'24 (2024)
  4. SPAGHETTI: a synthetic data generator for post-Covid electric vehicle usage A. Berkes, S. Keshav
    Energy Informatics, 7(1), 1-21. (2024)

Talks

  1. One Policy to Rule Them All: Towards Foundation Models for Building Control Workshop on Computational Optimization of Buildings at ICML 2025, Vancouver.
    July 2025
  2. Toward Adaptive and Generalisable Building Control: Graph-Based Policies and World Models Rolnick Lab Seminar at Mila (Qu´ebec Artificial Intelligence Institute), Montreal.
    May 2025
  3. Addressing challenges in building controls: toward more adaptive, data-efficient approaches Bengio Lab Seminar at Mila (Québec Artificial Intelligence Institute), Montreal.
    March 2025
  4. Using AI to decarbonise buildings Women in CS seminar, Dept. of Computer Science & Technology University of Cambridge.
    February 2025
  5. SOPEVS: Sizing and Operation of PV-EV-Integrated Modern Homes The 15th ACM International Conference on Future and Sustainable Energy Systems, Singapore.
    June 2024
  6. SOPEVS: Sizing and Operation of PV-EV-Integrated Modern Homes Energy and Environment Group, University of Cambridge
    24th of May 2024
  7. Reimagining Residential Microgrids: Policy Implications of Bidirectional Electric Vehicles, Solar Energy and Work-from-Home Trends Energy and Environment Group, University of Cambridge
    16th of February 2024
  8. The Optimal Design of PV-EV Integrated Domestic Microgrids - A Post-Pandemic Perspective Energy and Environment Group, University of Cambridge
    24th of January 2024
  9. Designing PV-EV integrated Residential Microgrids in the post-COVID World Energy and Environment Group, University of Cambridge
    16th of June 2023

Poster

  1. HVAC-GRACE: Transferable Building Control via Graph Recurrent Neural Networks 42 nd International Conference on Machine Learning (ICML 2025) - CO-BUILD Workshop
    July 2025
    Download Poster
  2. HVAC-DPT: A Decision Pretrained Transformer for HVAC Control 38th Conference on Neural Information Processing Systems (NeurIPS 2024) - Tackling Climate Change with Machine Learning Workshop & Women in Machine Learning (WiML) Workshop
    December 2024
    Download Poster
  3. Decarbonising Residential Homes in the UK: Synergy Benefits of PV, EVs and Work-From-Home The Architecture of Green Energy Systems: The Underlying Problem and Its Challenges Poster Session, IMSI Institute, University of Chicago
    17th of June 2024
    Download Poster
CSNOW logo Momentum logo VIS logo

Extracurricular Experience

  1. UNECE Expert Group on Energy Systems
    Young Member (November 2024 - Present) Advisor in the Energy Systems Young Member Group of the United Nations Economic Comission for Europe (UNECE).
  2. COP 29
    University of Cambridge delegate (November 2024) I represented the University of Cambridge at the UN Climate Change Conference in Baku, Azerbaijan.
  3. women@CL
    Tech Event Chair (September 2024 - Present) Women@CL supports women in computing research at the University of Cambridge.
  4. Solarplexus
    Founder (Jan 2024 - Present) NGO i-Teams project in collaboration with the Centre for Global Equality & Happy Hearts Indonesia.
  5. Momentum
    Founding Member (September 2021 - Present) Connecting young entrepreneurs with tech and business backgrounds to create a network of like-minded individuals across European universities.
  6. CSNOW – Network of Women in Computer Science at ETH Zürich
    Committeee Member and Teaching Assistant (February 2021 - October 2022) CSNOW was founded in 1993 and is part of the Computer Science Department. Women are strongly underrepresented at the department with 15% (see statistics). Our goal is to draw attention to the problematic effects of this unequal gender distribution and to counteract them. To this end, we organise various events and support female students with offers such as the mentoring programme, career workshops, interview series and networking events.
  7. VIS - Verein der Informatik Studierenden an der ETH Zürich
    University Policy Committeee (March 2021 - October 2021) The official Association of Computer Science Students at ETH Zurich (VIS) was founded in 1984 and is a professional association of the Association of Students at ETH (VSETH). For the more than 1,900 Bachelor and Master students of Computer Science, Data Science and Computational Biology and Bioinformatics, the VIS is the first point of contact for events, excursions, support during their studies and representation vis-à-vis the Department of Computer Science at ETH Zurich.