Inspiration
Cloud AI compute costs companies over $250 billion a year and data centers emit about 300 million tons of CO₂ annually (IEA 2023; Statista 2024).
What it does
Suchi is an AI-powered scheduler that decides where to run jobs to minimize cost, CO₂ and water use.
How I built it
Using Anthropic Claude for reasoning and Groq’s LPUs for ultra efficient inference; Suchi can reduce the financial and environmental footprint of AI compute by 10-40% (This range reflects real-world variation in grid carbon intensity, data-center efficiency [PUE 1.3 -> 1.1] and water use [WUE ~1.9 -> 1.2 L/kWh], as documented in Google Cloud and Uptime Institute sustainability reports. The range is variable and could be more or less depending on multiple parameters).
Challenges I ran into
Integrating real sustainability data and multiple APIs under strict time limits was tough.
Accomplishments that I am proud of
For large AI teams, this could mean millions saved and a measurable drop in carbon.
What I learned
I learned how to design a multi-objective job scheduler!
What's next for Suchi
Add predictive scheduling using historical patterns and connect directly to Kubernetes or cloud APIs for real job dispatch.
