ML Data & Eval Engineer, Siri Agent Modeling

The frontier of AI isnt in the cloud anymore. Its in your pocket. Every model that runs on a users device depends on a lot of systems behind it: data pipelines that decide what it learns, infrastructure that trains and evaluates it at scale, and tooling that gets it efficiently onto Apple silicon.

At Apples Siri team, were building the next generation of agents and deploying them at the largest customer scale in the world. Were looking for generalist engineers who like working across the whole ML lifecycle. You might build a data pipeline one quarter and a training platform the next, wherever the biggest bottleneck is.

This role is for builders who want end-to-end ownership of the data and infrastructure behind frontier on-device ML, from raw data to silicon. As an ML Data & Eval Engineer, youll work closely with Siri on-device modeling, evaluation, and deployment teams, and with Apples ML platform, compute, framework, and privacy partners. Youll design, build, and run the systems that turn data into trained, evaluated, and deployable on-device models. Youll own your services and data products through every stage: conception, design, implementation, deployment, and maintenance.Your Playground:

Data as a Product: Build scalable pipelines for curating, filtering, deduplicating, augmenting, and versioning the data behind our LLMs and agents. This includes fine-tuning sets, preference data, tool-use trajectories, evaluation sets, and synthetic data generated by models. Evaluation at Scale: Build the systems that run large evaluation suites for agentic models, including sandboxed environments for tool use and multi-step tasks, so every checkpoint gets measured quickly and consistently. Path to Device: Build the pipelines that take models from training to Apple silicon, including conversion, compression and quantization workflows, on-device benchmarking, and regression tracking, so every candidate model can be checked against real device constraints. Velocity at Scale: Find and remove bottlenecks across the ML lifecycle, including data loading, experiment tracking, and artifact management, so modelers spend their time on models and not on plumbing. Quality & Privacy by Design: Build data quality checks, lineage, and governance tooling that keep our datasets and models trustworthy and consistent with Apples privacy principles.

Were a team that values velocity and pragmatism. We debate, we build, we measure, and we ship. You are an ideal candidate if youve ever been woken up by a failed pipeline or training job and fixed it so it never happens again. Strong computer science background in algorithms, data structures, and distributed systems design Strong proficiency in Python, plus hands-on experience with Docker, Kubernetes, and cloud infrastructure Experience building large-scale data processing pipelines (e.g. Spark, Ray) or distributed ML infrastructure Familiarity with the ML model lifecycle (training, evaluation, deployment) and frameworks such as PyTorch or JAX Experience building, deploying, and debugging production services and data products end to end Experience with agentic AI-assisted codingBachelors degree in Computer Science, Engineering, or a related discipline, or equivalent industry experience. 4+ years of industry experience building production data or ML infrastructure. Experience building data pipelines for LLMs, such synthetic data generation, LLM-as-judge filtering, or evaluation sets Experience with workflow orchestrators such as Airflow, Kubeflow, or Argo Experience with model conversion and optimization for on-device or edge deployment (e.g. Core ML, quantization, pruning) Experience with data quality, lineage, and dataset versioning tools, and with privacy-preserving data practices Experience building tools and services that optimize ML systems beyond model selection Collaborative, with experience driving large cross-team projects