A journey distributing millions of residential energy consumption simulations with Clojure - Mendoza

How a platform scaled residential energy modeling simulations to millions of properties using a distributed Clojure supervisor architecture.

In this talk we follow the evolution of a platform that handles the energy modeling simulations from tens of thousands up to a hundred million US residential properties over time without going bankrupt. The core physics modeling engine is made publicly available by NREL. The main approach for developing such a platform focuses on three key points. First, massively distributing the modeling engine with a Clojure host supervisor on each instance. Second, addressing the most pressing computing resource needs at different pipeline stages. And third, maintaining vigilant attention to cost-effective choices. All efforts and decisions were framed and shaped by business and market conditions, opportunity costs, available engineering and infrastructure resources. Expect not many code examples, but diagrams and battle anecdotes related to the problem domain and how Clojure helped to bring the platform to success. Biography Joshua Mendoza first came to know Clojure around 2011. After working with mainstream languages for many years and using Clojure in smaller and personal projects, he decided to stick to it and call it home. He has led teams of software developers and also spent way too much time working with servers and cloud providers. As a footnote, he enjoyed his time teaching at the university and loves traveling with his wife and learning human languages. Recorded Nov 14, 2025 in Charlotte, NC at Clojure/Conj 2025.