In this role, you will develop, debug, and maintain data-driven features of a large-scale batch focussed compute platform. You will:
* Own features end to end that turn job behavior into decisions the platform acts on — from the initial data analysis through the model or heuristic, the APIs, production serving, and the measurement that proves it worked
* Implement your own changes across the stack, from platform services and control plane paths down to node-level resource limits and configuration
* Design and run validation in production: shadow mode, staged rollouts, explicit success metrics, and a rollback plan
* Have full ownership for the decisions your models make. When a prediction misbehaves, diagnose it, bound the impact, and help fix the system that trusted it
* Write and review code, generate and review design documentation
* Engage directly with customers and partner teams on their compute needs: understand their workloads, analyze cluster utilization and capacity, and turn that analysis into demand forecasts, node pool configuration and quota decisions that balance customer demand against fleet efficiency
* Bring quantitative analysis to open questions across the team — sizing, planning and prioritization decisions where good data changes the answer
* Participate in software qualifications and rollouts to production clusters
* Participate in an on-call rotation where engineers respond to platform issues for same-day resolution
* Work with a wide range of software and hardware engineering teams across Apple to support their workflows or integrate their technology into our platform
* Hold yourself and others to a high quality standard expected of Apple products
Minimum Qualifications
Strong programming skills in a general-purpose language (Go, Python, C++, or similar), and demonstrated ability to design, debug, and test complex software systems
Working knowledge of distributed systems and operating system fundamentals
Experience instrumenting large-scale infrastructure and analyzing the data it produces — reasoning quantitatively about how CPU, memory, I/O or network consumption behaves and how it scales — and turning that analysis into a decision or a measurable improvement in production
Comfort reasoning quantitatively about production data — distributions and tails, not just averages — and about the risk a wrong estimate creates for a running workload
Track record of owning a project from an ambiguous problem statement through production
Strong communication and organizational skills, including the ability to work directly with customers and to make quantitative results legible and actionable for people who are not data specialists
BS in Computer Science / related fields and 3+ years of experience or MS with 1+ years of experience or PhD
Preferred Qualifications
Experience building or operating a large-scale compute platform with responsibility for its capacity, efficiency, or performance
Experience improving utilization on a production platform — right-sizing requests from historical usage, oversubscription, bin packing, co-locating batch alongside latency-sensitive work, or reclaiming unallocated capacity. Comparable work on Kubernetes VPA, on runtime or demand prediction feeding an HPC scheduler, or on an in-house equivalent is equally relevant
Experience characterizing workloads at fleet scale, for example grouping jobs into behavioral classes or building the telemetry to do so
Experience shipping a model or heuristic that made automated decisions in production, and owning the outcome
Experience with forecasting, regression, or uncertainty estimation applied to operational time series
Experience modelling how a system's resource consumption scales — for example projecting the network, storage or I/O demand created by growing a compute footprint — and using that to inform capacity or sizing decisions
Experience with Kubernetes / Slurm or a comparable batch scheduling system