Machine Learning Engineer, ML/GenAI Evaluation

Apple•Austin, TX

About The Position

Would you like to contribute to Machine Learning and Generative AI technologies? Are you passionate about measuring what matters and ensuring AI systems work reliably for everyone? Do you believe that rigorous evaluation — including holding models accountable to fairness standards — is what separates great ML from good ML? We truly believe it is! We are defining what exceptional looks like for machine learning across Wallet, Payments, and Commerce. As a Machine Learning Engineer specializing in Evaluation, you will establish the evaluation criteria, metrics frameworks, and quality standards that determine when models are ready to reach hundreds of millions of users. Your judgment shapes model quality and earns the confidence to ship. You'll work at the intersection of rigorous ML science and high-impact product decisions, collaborating closely with ML Engineering, Product, Privacy, and Legal teams. This unique opportunity puts you at the center of model quality — designing adversarial test strategies, surfacing failure modes before they reach users, and owning the sign-off process that ensures Apple's financial features meet the highest bar for accuracy, robustness, and reliability. The ideal candidate is a rigorous, curious ML practitioner who believes that how you measure a model is just as important as how you train it. You think critically about what metrics actually capture, know how models break in the real world, and hold quality standards others find uncomfortably high — including on dimensions like fairness. You will own the full evaluation lifecycle for ML models across Wallet features — designing test frameworks, adversarial corpora, and benchmarks that reflect the diversity of Apple's global user base, then making the final quality call before any model ships. Your findings directly shape model development priorities and product decisions at scale.

Requirements

M.S. in Machine Learning, Computer Science, Statistics, Applied Mathematics, or a related technical field strongly preferred.
Bachelor's degree with 7+ years hands-on experience in ML evaluation, model quality, or applied research will be considered
5+ years of hands-on ML experience, with deep expertise in model evaluation, offline metrics design, and behavioral testing
Strong track record designing evaluation frameworks for production ML systems — not just accuracy/F1, but precision-recall tradeoffs, calibration, fairness, and task-specific quality dimensions
Creative mindset with the ability to translate standard ML evaluation metrics (F1, AUC, etc.) into utility and user trust measures
Experience testing for distribution shift, out-of-distribution generalization, and temporal drift in real-world deployed models
Proven ability to construct adversarial test suites, aggressor scenarios, and edge-case corpora that surface model failure modes before they reach users
Experience with structured and semi-structured document understanding, OCR pipelines, or financial data extraction is a strong plus
Strong programming skills in Python; fluency with evaluation tooling, data pipelines, and experiment tracking (e.g., MLflow, W&B, or equivalent)
Excellent communication skills — ability to translate metric results into product-quality narratives for engineering and executive audiences
Experience owning model quality sign-off in a cross-functional launch process

Nice To Haves

PhD in Computer Science, Data Science, Statistics, AI/ML, or a related field.
Experience with Bayesian or causal graph-based approaches to data generation.
Experience with causal approaches to fairness evaluation — counterfactual fairness, causal Shapley values, or structural causal model–based bias auditing.
Experience evaluating models under privacy constraints or on-device inference settings is a plus.
Familiarity with confidence calibration techniques and uncertainty quantification a plus
Background in financial services, fintech, or consumer payment products

Responsibilities

Define evaluation criteria and quality metrics for ML models powering Wallet features
Design and maintain structured test sets covering the full diversity of real-world scenarios — varied document formats, distributions, languages, edge cases, and adversarial inputs.
Develop evaluation methodologies for robustness testing: distribution shift, out-of-distribution generalization, temporal drift, and aggressor scenarios
Own fairness evaluation end-to-end — define fairness metrics appropriate to each Wallet feature, build bias test suites across protected attributes and user populations, measure disparate performance across subgroups, and gate model launches on fairness criteria with the same rigor as other conventional metrics.
Build user persona–stratified benchmarks that reflect the breadth of Wallet's global user population across spending patterns, locales, and document types
Evaluate generative and agentic model outputs — assessing hallucination rates, faithfulness, and groundedness using LLM-as-a-judge frameworks, human evaluation protocols, and prompt regression testing
Own model quality sign-off — establish the launch criteria, run final evaluations, and make the call on model readiness before any feature ships
Synthesize evaluation results into clear, actionable insights that guide model development priorities and product decisions
Partner with ML engineers and Quality engineers to identify failure modes early in the development cycle and close the loop between evaluation findings and model improvements
Establish and evangelize evaluation best practices across the Wallet ML team, raising the quality bar for how models are tested, monitored, and maintained post-launch