SUSTAIN

Weekly Energy Gap
Routing Intelligence
Metrics and KPIs

HackEurope Logo

Inspiration

Hours of painful MLOps and the growing guilt of knowing every GPU training run is quietly destroying the planet. AI compute is expensive, we're talking real dollars and we're broke students. Like most entrepreneurs and SMEs trying to build ML systems, we're cost-sensitive and environmentally conscious, but the tooling rarely reflects that. We wanted to change that.

What it does

A carbon-aware job scheduler for ML workloads. You submit a GitHub link and your scheduling preferences (max wait time, country allow-list), and the platform handles the rest forecasting energy supply-demand gaps across the EU using ML models trained on historical environmental data and live weather signals, then routing your job to the cheapest green window. Carbon monitoring is injected directly into the runtime, and the job deploys automatically on AWS, Azure, or GCP using spot infrastructure to keep costs low.

How We Built It

Multi-layered: Redis queues for job ingestion and enrichment, Claude Code as an AI agent to restructure code and estimate runtime, Railway for ML model orchestration, and Docker for bundling carbon hooks with the compute job. Individual forecasting models (one per country), each with 12h horizon gap predictions in MW.

Challenges

The integration surface was enormous: data engineering, MLOps, signal processing, cloud deployment, microservice orchestration, all in under 24 hours. Sourcing clean energy market data per country was painful. Red Hat OpenShift API integration was particularly rough.