Job description

What makes our work unique is we have many research opportunities in real-world use cases in multimodal space. Our goal is to develop targeted, task-specific capabilities within our models across a broad range of areas, including video understanding, reasoning, instruction following, and agentic tasks. The role involves contributing to every stage of the post-training process: Formulate hypotheses for a use case Define data collection and annotation Build evaluations of model capabilities Apply SFT, advanced reward modeling, RL techniques, and distillation to develop and improve the post-training recipe Execute experiments, validate results, and communicate findings to the broader team If you're passionate about doing research that ships in Apple's legendary products, this is the team for you. MS in Computer Science, Machine Learning, or a related field, or equivalent practical experience Minimum 4+ years of research experience in Multi-Modal LLM training and evaluation Proven track record in multimodal SFT plus at least one of distillation, preference optimization, or online RL, backed by shipped models or rigorous experiments Strong programming skills in Python and at least one deep learning, framework, such as JAX or PyTorch. Ability to build intuition from a problem statement and translate it into dataset requirements, neural network design, and loss functions. PhD in Computer Science, Machine Learning, or a related field, or equivalent practical experience Experience in GRPO, DPO, Vision-OPD, and visual-token compression Publications in top-tier ML conferences (e.g., NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV) Bonus: Experience in quantization, on-device models and SLM (Small Language Model)

Requirements

Must-haves

The role involves contributing to every stage of the post-training process:
Formulate hypotheses for a use case
Define data collection and annotation
Build evaluations of model capabilities
Apply SFT, advanced reward modeling, RL techniques, and distillation to develop and improve the post-training recipe
Execute experiments, validate results, and communicate findings to the broader team
If you're passionate about doing research that ships in Apple's legendary products, this is the team for you.
MS in Computer Science, Machine Learning, or a related field, or equivalent practical experience
Minimum 4+ years of research experience in Multi-Modal LLM training and evaluation
Proven track record in multimodal SFT plus at least one of distillation, preference optimization, or online RL, backed by shipped models or rigorous experiments
Strong programming skills in Python and at least one deep learning, framework, such as JAX or PyTorch.
Ability to build intuition from a problem statement and translate it into dataset requirements, neural network design, and loss functions.
PhD in Computer Science, Machine Learning, or a related field, or equivalent practical experience
Experience in GRPO, DPO, Vision-OPD, and visual-token compression
Publications in top-tier ML conferences (e.g., NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV)

Nice-to-haves

Bonus: Experience in quantization, on-device models and SLM (Small Language Model)

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

machine learningllmpythonml

Platform: apple
Effort: medium

Benefits

Benefits detected in the listing.

Vision