.png&w=256&q=75)
🤓 Latest Submissions
GreenQueue
GreenQueue is an AI compute scheduler that decides when a workload should actually run, based on real electricity grid conditions — not just when it was submitted. The problem: most compute runs the instant it's requested, even though grid carbon intensity swings significantly across the day. The same job can have a larger footprint at 3pm during peak fossil-fuel generation than at 3am when renewables dominate. Almost no scheduler accounts for this. GreenQueue does. Every job carries a priority. Urgent jobs dispatch immediately — no exceptions, no compromise on latency. Flexible jobs are checked against a live carbon-intensity threshold: above it, the job waits; below it, it runs. A background process continuously re-evaluates conditions, so waiting jobs dispatch automatically the moment the grid clears — no manual intervention. A safety valve forces dispatch after a max delay so nothing waits forever. For every flexible job GreenQueue delays and later dispatches, it records the real difference between carbon intensity at submission versus dispatch — genuine avoided carbon, not an estimate. Aggregated, this produces concrete metrics: carbon avoided, cost saved, jobs optimized, and flexible compute-hours shifted into cleaner windows. Architecturally, GreenQueue is a scroll-driven 3D frontend (React, Three.js, React Three Fiber) visualizing the scheduling story — workloads arriving, urgent dispatch, flexible waiting, the grid changing, a clean window opening, execution, impact — above a real dashboard (grid trend, job queue, scheduler pipeline, impact metrics). The backend is a FastAPI service implementing the actual scheduling engine, exposing one snapshot endpoint so every visual reads identical, real state. Built for the AI Infra Summit Hackathon 2026's Compute track: a genuine systems-level optimization, not another app layered on an existing model — visuals and logic both working end-to-end.
16 Sep 2026
