Sr. ML Production Model Automation Engineer, Siri Speech

Apple Apple · Big Tech · Cupertino, CA +1 · Machine Learning and AI

This role focuses on automating the production model lifecycle for Siri's speech and audio features, which are powered by multimodal, on-device AI. The engineer will build and operate agent-based automation pipelines for ML model training, iteration, staging, rollout, and deprecation, including SFT, LoRA, and RL phases. The work involves developing multi-agent workflows for evaluation, triage, and root cause analysis, and owning the launch tooling for training jobs.

What you'd actually do

  1. Own the end-to-end model lifecycle building model pipelines, integrating with other Apple frameworks to enable rapid model iteration, staging promotion, production rollout and deprecation.
  2. Design and operate agent-based automation pipelines for ML models where agents own decision logic at each gate and humans approve only at defined escalation points
  3. Develop multi-agent workflows using LLM-native tooling for on-device evaluation, regression triage, release readiness decisions, and automated root cause analysis.
  4. Own the launch tooling to build and improve the shell scripts and CLI commands that turn a config-name and a dataset into a running training job — across SFT, LoRA adapter, and RL phases.

Skills

Required

  • Python
  • Bash
  • Machine Learning Operations
  • Cloud ML platforms (GCP TPU, AWS GPU clusters, Kubernetes-backed training infra)
  • ML training lifecycle (data preprocessing, distributed training, checkpoint formats, multi-slice/multi-region)
  • Infrastructure-as-code
  • CLI tool design
  • Observability
  • Reliability

Nice to have

  • JAX
  • XLA
  • Large-model training stacks
  • Multi-slice TPU training
  • Cross-region GCS / S3-compatible storage
  • MLOps tools (model registries, feature stores, experiment trackers, reward-model serving)
  • Apple Access Manager
  • AWS IAM at scale
  • Claude Code / agent skills
  • Runbooks
  • LLM-assisted developer tooling

What the JD emphasized

  • 5+ years experience in Machine Learning Operations
  • Production experience with one or more cloud ML platforms
  • Familiarity with the ML training lifecycle
  • Experience with infrastructure-as-code, CLI tool design, and developer ergonomics
  • Bias toward observability and reliability

Other signals

  • building products for voice, dictation and other audio products
  • multimodal models that power Siri on-device speech features
  • next generation of audio experiences across our platforms
  • train models, iterate on data mixtures
  • stack supervised fine-tuning, LoRA adapter training, and reinforcement learning into pipelines
  • turn the operational substrate underneath foundation model training into a reliable, observable, self-serve system
  • Design and operate agent-based automation pipelines for ML models where agents own decision logic at each gate and humans approve only at defined escalation points
  • Develop multi-agent workflows using LLM-native tooling for on-device evaluation, regression triage, release decisions, and automated root cause analysis
  • Own the launch tooling to build and improve the shell scripts and CLI commands that turn a config-name and a dataset into a running training job — across SFT, LoRA adapter, and RL phases