Research Engineer / Scientist

Apple•Cupertino, CA

About The Position

Apple is seeking a technical leader to drive training data and model evaluation across the foundation model training lifecycle, from pre-training and mid-training through post-training. The role combines machine learning research, model evaluation, and distributed data systems, supporting high-performance server models and efficient on-device models designed for Apple silicon and private, personal use. The position shapes strategy for training data and evaluation, ensuring data is high-quality, compliant with privacy and policy requirements, and targeted through curation and synthesis based on product feedback. It also drives evaluation that guides model training and predicts real user experience. The role involves balancing data, evaluation, and training priorities and bringing findings into decisions about capabilities, data mixtures, and release readiness. The successful candidate partners with model training leads, product teams, and security, legal, and compliance teams, and leads through technical depth and influence. Depending on the candidate’s background, the role may include managing a team. The role’s scope includes production model strategy; connecting evaluation, data, and training; evaluation; data systems and curation; and technical and team leadership. Responsibilities span strategy and standards, data and evaluation research, data infrastructure, governance, and cross-functional leadership. The candidate’s focus will depend on their background and may grow over time.

Requirements

PhD and 5+ years, or MS and 8+ years, of experience in machine learning or data infrastructure, including leading technical work at scale.
Deep machine learning expertise in LLMs or multimodal foundation models, with depth in either evaluation methodology or large-scale training data systems and curation, and working fluency in the other.
Strong fundamentals in statistics and experimental design, with the ability and credibility to review designs and experiments in detail.
Track record of training data or evaluation work on large models that measurably improved model quality or changed how models are built.
Technical leadership across teams, including setting direction, mentoring senior engineers or researchers, and driving adoption of systems, benchmarks, or standards beyond one’s own team.
Ability to influence senior leaders and cross-functional partners and turn complex results into clear decisions.
Hands-on proficiency in Python and an ML framework such as PyTorch or JAX.

Nice To Haves

PhD in a related field.
Experience across pre-training, mid-training, and post-training data, including data ablations and scaling-law studies, and experience with non-text modalities such as images, audio, and video, and source code at scale.
Experience designing standardized LLM benchmarks, LLM-as-a-judge frameworks, contamination detection and decontamination, or human evaluation programs with annotation teams and vendors.
Experience architecting distributed data pipelines, such as Apache Beam or Spark, on major cloud platforms; strong systems programming skills; and experience building evaluation or data platforms adopted across multiple teams.
Experience with data provenance, licensing, and privacy-preserving practices for training data.
Experience managing a team of engineers or researchers.

Responsibilities

Shape training data and evaluation strategy across pre-training, mid-training, and post-training, aligned with the production model roadmap.
Help determine investment across data acquisition, curation, synthetic data, and evaluation coverage, including what to build, buy, or source from vendors.
Set standards for benchmark governance, release gating, contamination, and data documentation.
Translate evaluation failures into data decisions, determine capability gaps to address through acquisition, curation, or synthetic generation, and verify results.
Lead data ablations and scaling-law studies, and define how data-mixture decisions are made and validated.
Lead decontamination across pre-training, mid-training, and post-training data, including synthetic and vendor data; define standards, measure residual contamination, and protect held-out evaluation sets from overuse.
Define evaluation portfolios for reasoning, coding, knowledge, and multi-turn agentic workflows, and determine where to build benchmarks or adopt existing ones.
Lead model-based judging, including judge validation, versioning, and bias checks.
Build product-aligned evaluation that predicts user-perceived quality on Apple devices, and work with annotation teams and vendors to correlate automated metrics with human judgment.
Direct the architecture of high-throughput systems that acquire, process, and curate petabytes of text, code, and multimodal data across cloud and high-performance compute.
Set quality standards for algorithmic data curation, including semantic deduplication, heuristic and model-based quality filtering, and synthetic data generation.
Drive data provenance, licensing, privacy, and takedown practices with security, legal, and compliance teams.
Raise technical standards through design reviews, mentoring, and hiring, and drive adoption of shared systems and tools across teams.
Manage and grow a team if the role includes one.
Communicate findings, risks, and recommendations clearly to senior leadership and cross-functional partners.