Hello and welcome!
First, who am I? I specialize in taking AI systems from research concepts to production at scale. I close the gap between research and practice. My focus is primarily around modelling dynamical systems for interventions under uncertainty.
At BrainBox AI, leading a team of 12, I helped deploy thousands of models by standardizing how buildings are mapped, modelled and controlled. Our breadth differentiated us as a company that can scale quickly, saving our customers 15-25% of their utility bills within days of onboarding. BrainBox AI was acquired by Trane in January 2025.
At RailVision, I was brought in as a simulations expert. I rewrote our passenger and freight train physics simulator in a differentiable and compilable framework, improved the calibration process while making simulations 1000× faster. The accuracy and speedup opened up the door for continuous optimization saving customers 15-20% on fuel and improving safety.
These days, I've co-founded Aescia Health where we have built a medical platform that allows patients to stay in touch with the medical staff post-operation and allows the medical staff to monitor their healing and avoid costly readmissions. Our vision is that no patient should leave the hospital feeling alone. Health is a domain I've been curious about having an impact in, and with our trial in Australia, we're already seeing results.
My background: I hold a PhD in Building Engineering where my work was to model and optimize the operation of a net-zero energy building. I went from extracting the raw data, to building models, to then balance the energy consumption and production on-site. I also hold a Master's degree where my work was around energy storage using phase-change materials. Prior to that, I worked as an MEP engineer designing the HVAC and plumbing systems of all major hospitals in Montreal, Canada.
This site also serves as my blog where I sporadically publish some (half-baked) ideas. Other links are included in the sidebar. Thanks for stopping by. — Vasken Dermardiros
Availability
- Where
- Based in Barcelona. Hybrid in Barcelona preferred — I like being in a room with people — and fully remote works too.
- How much
- Part-time through full-time. The arrangement I'd like most is four days a week somewhere, with the remaining two on Aescia Health. Full-time for the right role.
- What
- Physics-informed machine learning on real systems: world models, building and HVAC optimisation, grid and energy systems, and adjacent physical-systems modelling. Fifteen years of dynamical systems is the throughline: that's the work I want.
- Where not
- I use LLMs and agents as tooling systems. I'm not looking to build chatbots or document-extraction products. I'm also looking for an established company rather than a seed-round early-stage start-up — I already have one of those.
Now
- 2025–
- CTO & Co-Founder, Aescia Health.Aescia Health is a start-up providing medical discharged patients a structured recovery journey while allowing hospital staff to monitor their progress and reduce costly readmissions. Every technical decision is mine: infrastructure, backend, frontend, database, the AI layer — LLM-RAG over discharge summaries into personalised care plans, agents, and the answer parsing that makes structured clinical Q&A feel like a conversation — plus SOC 2 and HIPAA compliance and the patient- and clinician-facing apps. Clinical trial underway at Royal Prince Alfred Hospital, Sydney. Scaling through the District 3 accelerator in Montreal.
Previously
- 2023–25
- Senior ML & Simulation Engineer, RailVision Analytics.RailVision Analytics is a seed stage start-up with the goal to enable self-driving efficient trains. Train physics simulator rewritten in JAX (1000× faster, better accuracy); real-time driving recommendations worth 20% fuel on freight, 15% on passenger rail; validated over four on-train runs.
- 2022–23
- Senior ML Engineer, BrainBox AI (now part of Trane).BrainBox AI is a ClimaTech company dedicated to reducing energy and carbon emissions of retail and office buildings through HVAC optimization. Cut the training data needed per building by 80% through scientific ML, taking customer onboarding from weeks to days. Built the digital-twin framework behind rapid prototyping of control strategies, and LLM agents for parsing building specifications. Led the technical due diligence supporting multi-million dollar funding rounds and partnerships. Two US/Canada patents as main inventor and four published articles. Delivered the ML platform that made the Trane acquisition happen.
- 2020–22
- Engineering Manager, AI, BrainBox AI. Scaled the ML infrastructure from dozens of models to 2000+ running in production, and architected the end-to-end pipelines for HVAC control and anomaly detection behind them. Automated model lifecycle management until maintenance overhead disappeared. Grew and led a team of 12 engineers, promoting 3 of them into management.
- 2019–20
- Machine Learning Engineer, BrainBox AI. Built the temperature prediction framework deployed across 50+ commercial buildings, and the optimisation and RL-based control delivering 15–25% energy savings on real HVAC equipment.
- 2013–21
- Research Scientist & TA, Concordia University. Partnerships with Hydro-Québec LTE-IREQ, Régulvar and NRC-CanmetENERGY. Most of the publications below.
- 2011–13
- MEP Engineer, Bouthillette, Parizeau et Associés.BPA provides mechanical, electrical, structural, and sustainability engineering services. Mechanical design for healthcare and commercial buildings; automated CAD/BIM workflows.
Education
- 2015–20
- PhD, Building Engineering, Concordia University. Supervised by Andreas K. Athienitis and Scott Bucking. Defence graded outstanding. Holder of the NSERC Alexander Graham Bell Canada Graduate Scholarship (CGS D3) — one of three at Concordia that year.
- 2013–15
- MASc, Building Engineering, Concordia University.
- 2007–11
- BEng, Building Engineering, Concordia University. Honoured with Distinction.
Stack
- ML & scientific
- Python, JAX, PyTorch, TensorFlow, Keras, scikit-learn, CVXPY, pandas, Julia, and, of course, Excel.Yes, Excel. It's where a surprising amount of real engineering data lives, you can't beat a quick pivot table with filters to help decide if we need a deep analysis or not.
- Production
- AWS (ECS, ECR, S3, CodeArtifact), GCP, FastAPI, Docker, Kubernetes, Airflow, Supabase, PostgreSQL, MongoDB, InfluxDB, Grafana, ClearML, MLflow, Vercel, Terraform, Terragrunt.
- Agentic
- Pydantic Agents, Langfuse, llama.cpp, Hermes, RAG, Claude Code.Actively experimenting with small models that summarise and call tools well — a model doesn't need to have memorised Wikipedia if it can search it, and doesn't need to write code if it can hand that to something that does.
- Full-stack
- TypeScript/JavaScript, React, Node.js, HTML/CSS.
- Development
- Git, Linux, Vim, Bash, C/C++ (embedded systems).
- Languages
- English (native), French (C1, full professional proficiency), Spanish and German (A1, working towards A2), Armenian (native, mother tongue).
Publications
Theses
Dermardiros, V. (2021). Data-Driven Optimized Operation of Buildings with Intermittent Renewables and Application to a Net-Zero Energy Library. PhD dissertation, Concordia University. Link.
Dermardiros, V. (2015). Modelling and experimental evaluation of an active thermal energy storage system with phase-change materials for model-based control. Master's thesis, Concordia University. Link.
Journals
Mtibaa, F., Nguyen, K.K., Dermardiros, V., Cheriet, M. (2021). Context-Aware Model Predictive Control Framework for Multi-Zone Buildings. Journal of Building Engineering, Volume 42, October 2021, 102340. Link.
Dermardiros, V., Buonomano, A., Athienitis, A.K., Bucking, S. (2020, Rework). Sampling-Based Model Predictive Controls Application on a Net-Zero Energy Library. Science and Technology for the Built Environment.
Dermardiros, V., Bucking, S., Athienitis, A.K., Buonomano, A. (2020, Rework). A Methodology for Development of Grey Box Low-Order Models for Building Power and Thermal State Forecasting. Energy and Buildings.
Morovat, N., Candanedo, J.A., Athienitis, A.K., Dermardiros, V. (2019). Simulation and performance analysis of an active PCM-heat exchanger intended for building operation optimization. Energy and Buildings. Link.
Dermardiros, V., Athienitis, A.K., Bucking, S. (2019). Energy performance, comfort and lessons learned from an institutional building designed for net-zero energy. ASHRAE Transactions 125, Part 1.
Papachristou, A.C., Vallianos, C.A., Dermardiros, V., Athienitis, A.K., Candanedo, J.A. (2018). A numerical and experimental study of a simple model-based predictive control strategy in a perimeter zone with phase-change material. Science and Technology for the Built Environment. Link.
Bucking, S., Dermardiros, V. (2017). Distributed evolutionary algorithm for co-optimization of building and district systems for early community energy masterplanning. Applied Soft Computing Journal, Volume 63 (Feb 2018): 14-22. Link.
Dermardiros, V., Chen, Y., Daoud, A., Athienitis, A.K. (2015). Development of reduced order thermal models of building-integrated active PCM-TES. ASHRAE Transactions 2016, Volume 122, Part 1: 267-277. Link.
Bastien, D., Dermardiros, V., Athienitis, A.K. (2015). Development of a new control strategy for improving the operation of multiple shades in a solarium. Solar Energy. Volume 122 (Dec 2015): 277-292. Link.
Conferences
Mtibaa, F., Nguyen, K.K., Dermardiros, V., Cheriet, M. (2021). Online Genetic-Algorithm-based Model Predictive Control Framework for Multi-Zone Buildings. 2021 European Control Conference, ECC 2021, 1011-1017.
Amara, F., Dermardiros, V., Athienitis, A.K., Buonomano, A. (2020). Energy Flexibility Modelling and Implementation for an Institutional Net-zero Energy Solar Building and Design Application. 2020 Summer Study on Energy Efficiency in Buildings. ACEEE, Pacific Grove, CA.
Amara, F., Dermardiros, V., Athienitis, A.K. (2019). Energy flexibility for an institutional building with integrated solar system: Case study analysis. 2019 IEEE Electrical Power and Energy Conference (EPEC). Montreal, Canada.
Dermardiros, V., Bucking, S., Athienitis, A.K. (2019). A simplified building controls environment with a reinforcement learning application. IBPSA Building Simulation 2019, Rome, Italy. Repo.
Dermardiros, V., Vallianos, C., Athienitis, A.K., Bucking, S. (2017). Model-based control of a hydronic radiant slab for peak load reduction. IBPSA Building Simulation 2017, San Francisco, CA.
Avagliano, G., Buonomano, A., Cellura, M., Dermardiros, V., Guarino, F., Palombo, A. (2017). Buildings integrated phase-change materials: modelling and validation of a novel tool for the energy performance analysis. BSA 2017, Bolzano, Italy.
Dermardiros, V., Athienitis, A.K. (2016). Building-Integrated PCM-TES for peak load reduction. eSim 2016. Link.
Bucking, S., Dermardiros, V., Athienitis, A.K. (2016). The effect of hourly primary energy factors on optimal net-zero energy building design. eSim 2016. Link.
Dermardiros, V., Athienitis, A.K. (2015). Comparison of PCM-active thermal storage systems integrated in building enclosures. 10th Conference on Advanced Building Skins, Bern, Switzerland. Link.
Dermardiros, V., Chen, Y., Athienitis, A.K. (2015). Modelling of an active PCM thermal energy storage for control applications. Energy Procedia, IBPC 2015. Volume 78 (Nov 2015): 1690-1695. Link.
Kapsis, K., Dermardiros, V., Athienitis, A.K. (2015). Daylight performance of perimeter office façades utilizing semi-transparent photovoltaic windows: a simulation study. Energy Procedia, IBPC 2015. Volume 78 (Nov 2015): 334-339. Link.
Guarino, F., Dermardiros, V., Chen, Y., Rao, J., Athienitis, A.K., Cellura, M., Mistretta, M. (2015). Phase-change material (PCM) thermal energy storage in buildings: experimental study and applications. Energy Procedia, SHC 2014. Volume 70 (May 2015): 219-228. Link.
Posters & non-refereed
Amara, F., Dermardiros, V., Athienitis, A.K. (2019). Energy Flexibility for an Institutional Net-Zero Energy Building in a Cold Climate (Poster). 2019 IEEE EPEC, Montreal, Canada.
Dermardiros, V., Vallianos, C., Dumoulin, R., Kapsis, K., Athienitis, A.K. (2016). On Demand Side Management and Grid Interaction for Institutional Net-Zero Energy Buildings in Canada (Poster). 7th International Conference on Integration of Renewable and Distributed Energy Resources, Niagara Falls, Canada.
Dermardiros, V. (2014). Model predictive control of building-integrated thermal storage based on phase-change materials (Poster). 3rd Annual General Meeting of the NSERC-SNEBRN research network, Montreal, Canada.
Patents
Two US/Canada patents filed as main inventor, on building control and HVAC optimisation, during my time at BrainBox AI.
Talks
Guest lecture, “Towards large scale deployment of model-based controls” (March 2022). Physics-informed models and model-based predictive control.