Data Scientist - Intelligent Experiences, ML

Summary

Do you believe machine learning and AI can change the world? We do. We're the ML Data team within Apple's Intelligent System Experience (ISE) group, and we build the high-quality, high-scale datasets that train the ML features across many Apple products — iPhone, iPad, Mac, Apple Watch, AirPods and the newly released HomeView. Our work sits behind some of the most visible and personal experiences Apple ships: the wallpaper on your Lock Screen, personalized stickers in iMessage, on-device Genmoji and other Apple Intelligence generative models, and models that recognize faces and build memories of the people you love in Photos. We're looking for a data scientist who's passionate about ML data and excited to shape the training data behind Apple's next generation of multi-modal features. The ideal candidate thrives in a fast-paced environment, brings rigorous scientific reasoning and owns problems end-to-end. Join us and you’ll shape shipping products from your first week, working with petabytes of multi-modal data.

Minimum Qualifications

Bachelor's or Master's degree in Computer Science, Machine Learning, Data Science, or a related engineering field 5+ years of hands-on experience applying NLP, computer vision, or multimodal machine learning to production systems, including experience with large-scale data pipelines

Preferred Qualifications

Experience prompting, fine-tuning, and evaluating LLMs or vision-language/diffusion models for specific downstream tasks Experience with parameter-efficient fine-tuning (LoRA/QLoRA), preference optimization (RLHF, DPO), RAG and retrieval-based data curation Experience designing, evaluating and productionizing data science algorithms used for data curation, data filtering, deduplication and data down-selection. Experience designing controlled experiments or ablation studies to understand effect of data on model quality Experience building agentic or multi-agent, tool-using LLM systems and human-in-the-loop workflows, calibrating judges against human labels, and mitigating failure modes Strong software engineering skills, with 5+ years in Python, workflow orchestration (Airflow), cloud platforms (AWS, GCP), distributed data processing frameworks (Spark, Ray, or similar) at large scale Publication record or open-source contributions in NLP, computer vision, or generative AI is a plus Familiarity with privacy-preserving data techniques (e.g., differential privacy, on-device data constraints) is a plus

Description

You will design and operate the end-to-end data lifecycle that supports training, fine-tuning, and evaluation of large language models and diffusion models. Day to day, that means building distributed pipelines that clean, deduplicate, and enrich massive multi-modal datasets; standing up synthetic data generation pipelines with LLM-based generators, validators, and judges; and engineering agentic workflows — multi-step, tool-using LLM systems — that scale human judgment across far more data than a human team could review directly.

Responsibilities

Design synthetic data generation pipelines — prompt/context engineering, generator/critic loops, diversity and contamination controls and prove the data's impact on model quality through controlled ablations Engineer agentic workflows (multi-step, tool-using, multi-agent LLM systems) or LLM-based flows (RAG, fine tuning) for curation, annotation, labeling QA, and evaluation, with human-in-the-loop checkpoints where they matter Define and run quality evaluation for synthetic assets: perceptual and reference-based metrics alongside VLM judges and human preference studies Fine-tune and evaluate LLMs and diffusion models for specific downstream tasks; run controlled ablations that isolate the contribution of each data intervention Partner cross-functionally with modeling, product, and privacy/legal teams to translate data tradeoffs into decisions, and identify strategic opportunities to improve our data capabilities Track and evaluate emerging techniques in data curation and generative AI, and make evidence-based recommendations about what to adopt