As a Machine Learning Engineer focusing on on-device LLMs and adapter technologies, you will tackle some of the most challenging problems in modern applied machine learning: delivering state-of-the-art agentic capabilities, reasoning, and task completion under strict on-device compute, memory, latency, and context constraints.

You will lead the end-to-end lifecycle of on-device LLM adapters and sub-models—ranging from training recipe formulation, data curation, and architecture tuning to rigorous metric evaluation, error analysis, and on-device runtime optimization. You will work closely with cross-functional teams including Foundation Model teams and Product teams to scale Apple Intelligence across diverse languages and domains.

Minimum Qualifications

5+ years of hands-on industry experience building, training, and shipping production-grade deep learning and NLP systems at scale.
Deep Hands-on LLM Experience: Proven track record in training and fine-tuning Large Language Models (e.g., SFT, DPO/RLHF, PEFT/LoRA). Strong intuitive understanding of model behavior, loss dynamics, and scaling laws.
Advanced Engineering Skills: Exceptional coding skills in Python and proficiency with modern deep learning frameworks (PyTorch, DeepSpeed, Megatron-LM, Hugging Face, vLLM). Familiarity with high-performance C++ or Swift runtime integration is a plus.
Data & Evaluation Rigor: Solid experience designing domain-specific datasets, synthetic data generation pipelines, automated evaluation harnesses, and conducting systematic root-cause failure analysis.
Collaboration & Execution: Strong ownership, cross-functional communication, and project management skills; ability to navigate complex engineering constraints and deliver against tight product deadlines.

Preferred Qualifications

On-Device / Edge Deployment Experience: Deep understanding of edge-AI constraints and deployment pipelines (e.g., CoreML, ONNX, TensorRT-LLM, Apple Neural Engine / Metal optimization).
Agentic Modeling & Tool-Calling: Direct experience building agentic workflows, multi-turn reasoning, and function/tool-calling capabilities in compact/dense models (e.g., 2B–9B parameter class).
Proven research track record via publications in top-tier conferences (NeurIPS, ICML, ICLR, ACL, EMNLP) or demonstrated history of shipping industry-leading GenAI products.