Congrats!! Loved this team, the problem they solved, and what they build. 🚀
This past weekend, I won 1st place at Yale University’s annual hackathon for the Zed prize track! I worked with some amazing teammates (Stephanie M., Aslan Wang, David Zhang ) to create EcoRouter, a model router that selects the most environmentally friendly LLM for your task, without sacrificing performance. In our limited timeframe, we were able to show that some configurations of 11x larger AI models were actually more efficient than their smaller model counterparts. EcoRouter used a custom fork of Andrej Karpathy's autoresearch that performed over 6,000 inference experiments on open source models in under 24 hours using 2x NVIDIA DGX Spark computers. Our custom fork used a 67-channel sensor logging system to optimize for environmental impacts and performance. We then created linear regressions and 3D energy models to benchmark other LLMs against. This data was used in a custom Zed MCP application written in Rust to help users find LLMs that meet their environmental and performance needs. Thank you to the YHack and Zed Industries teams for an amazing weekend!