We won Gold at the Global AI Challenge 2025 for Building E&M Facilities

Thrilled to share that our team received the Gold Award at the Academic Track at the Global AI Challenge held in Hong Kong this August! 🇭🇰

The challenge?
👉 Predict 30 days ahead of cooling load for several commercial buildings — including those with limited historical data.
A tough task, especially in real-world conditions where data is sparse, and model explainability and robustness matter.

💡 Instead of relying on deep neural nets, massive training pipelines, or black-box models, we took a different path.

🚀 We developed the Adaptive Similar Day (ASD) method — a training-free, interpretable, and blazingly fast approach that:

🔹 Uses only air temperature
🔹 Finds and averages historically similar days
🔹 Applies a temperature-aware scaling adjustment

No training. No backprop. No heavy ML packages.
A Top5 result. 5000 USD prize money.

We’re proud to show that simple + smart can sometimes beat complex + opaque, and that classic ideas like similarity-based modeling still have a lot to offer when thoughtfully adapted.

Big thanks to the organizers, the judging panel, and the EMSD for an incredible experience and hospitality in Hong Kong