Senior Ml / Devops Engineer – Ai Scheduling & Optimisation

1 day ago Be among the first 25 applicantsDirect message the job poster from Project OhmFounder of Project Ohm, 2it, Duvtail, Investor, Connector, occasionally prone to an outburst on philosophy.Why Project Ohm?AI's energy appetite is exploding just as renewable power is being curtailed and wasted.

Project Ohm flips that equation : we run high-performance AI workloads next to stranded clean energy that would otherwise go unused.

Our orchestration platform dynamically schedules GPU jobs to follow real-time price and renewable availability—cutting costs and emissions at scale.Backed by Australian and international investors and selected to represent Australia at the 2025 UN AI for Good Summit, we're now hiring the brainpower to build the optimisation engine at the heart of our vision.The ChallengeDesign and deliver a rules-based, energy-aware scheduler that places thousands of containerised AI jobs across a decentralised fleet of GPU nodes—while reacting in milliseconds to wholesale market signals, network constraints, and SLA tiers.Why this role mattersProject Ohm's platform is built around a hub-and-spoke "islands" architecture : many autonomous GPU pods (Kubernetes clusters) positioned at renewable-rich sites, all steered by a central control plane.

Your code will power the Custom Rule / Algorithm Engine that decides—every few seconds—which pod gets which job based on energy price, carbon intensity, GPU type and SLA.What you'll tackleDesign & ship the scheduling coreExtend our Python / Go engine to drive Kueue + MultiKueue for gang-scheduling thousands of AI / HPC jobs across clusters.Encode cost- / carbon-aware heuristics and reinforcement-learning loops that react to live WEM / AEMO price feeds.Orchestrate at planet scaleWork with Rafay KOP to template fleet-wide blueprints, automate upgrades and enforce policy guard-rails.Leverage Pulumi IaC (wrapping Terraform) so infrastructure, network (via Netris) and app logic live in one repo.Keep data moving, not waitingOptimise Dragonfly P2P and NVIDIA AIStore caches so containers and datasets hit the GPUs in seconds, not minutes.Make the network disappearExploit Cilium eBPF + Cluster Mesh to link clusters, enforce zero-trust L3–L7 policy and visualise flows with Hubble / Tetragon.Measure what mattersInstrument utilisation, queue latency, $ / MWh saved and tCO2e avoided; surface it in Grafana & customer portals.About you6+ years building distributed systems, including Kubernetes-native scheduling or OR / ML optimisation.Deep hands-on with at least three of : Kueue / MultiKueue, Slurm, Kubeflow Trainer, Argo CD, GPU Operators, Pulsar / Kafka, Ray, Spark.Fluent in Python and Go; comfortable turning Jupyter notebooks into production micro-services.Strong DevOps chops : GitHub Actions, container-first CI / CD, observability stacks, chaos engineering.You think latency, bandwidth and egress dollars before writing code, and can explain eBPF to a junior in five minutes.Bonus points for : energy-market modelling, e-mobility or micro-grid optimisation, eXtended Berkeley Packet Filter wizardry, or having run infra in remote / harsh environments.What we offerMission with leverage – every algorithmic improvement cuts both carbon and cost for AI workloads at scale.Early-team equity – help define the platform and share in its upside.Remote flexibility – fly-in sprints to Perth's renewables-powered HQ when it adds value.Growth budget – pick the GPU workstation, conference or course that accelerates you.Founder access – report directly to the CEO; shape roadmap, hiring and open-source engagement.How to applyThis role is open to applicants located anywhere in Australia.Send your CV / GitHub plus