About The Position

The Multimodal Intelligence Team is building foundation models for Apple experiences. The AI Research Scientist will advance model architectures, pre-training methods, and distillation techniques to make multimodal models practical across the Apple ecosystem. Research spans the foundation-model lifecycle, including architecture, pre-training objectives, data mixtures, optimization, scaling, distillation, and evaluation. The work addresses how to combine broad intelligence with the memory, latency, energy, and privacy requirements of on-device deployment, and may lead to technologies for Apple products, publications, and open-sourced research artifacts, evaluations, tools, or selected models. The role investigates how multimodal foundation models should be designed, trained, and distilled. Responsibilities include developing and evaluating architectures, pre-training objectives, data strategies, optimization methods, and teacher–student learning techniques. A central focus is co-developing frontier and efficient models for Apple silicon and on-device intelligence, including architectures designed to distill effectively and methods that preserve reasoning, multimodal understanding, and instruction following under constrained model capacity. Deployment constraints are considered throughout research, from early architecture and pre-training experiments through distillation and final evaluation. Potential research areas include dense, recurrent, state-space, mixture-of-experts, and hybrid architectures; multimodal pre-training across language, images, video, audio, and sensor-derived representations; compute-optimal scaling; architecture and algorithm co-design for efficient inference on Apple silicon; and offline or on-policy distillation using teacher-generated data, logits, representations, rationales, and other supervision signals.

Requirements

  • Hands-on experience designing, implementing, and running large-scale pre-training experiments for large language models.
  • Experience with LLM pre-training topics such as model architecture, training objectives, data mixtures, tokenization, curricula, scaling, and optimization.
  • Strong proficiency with modern deep learning frameworks such as PyTorch or JAX and distributed training systems.
  • Experience evaluating pre-trained models across language understanding, reasoning, instruction following, or multimodal capabilities.
  • Strong understanding of transformer-based architectures and current approaches to efficient or scalable foundation-model training.
  • Master’s degree or equivalent practical experience in machine learning, computer science, or a related technical field.
  • A minimum of 6 years of relevant experience.

Nice To Haves

  • Experience contributing to major foundation-model pre-training efforts or leading architecture experiments that influenced a large training run.
  • Research contributions in model architecture, scaling laws, multimodal pre-training, optimization, efficient attention, mixture-of-experts, state-space models, or related areas.
  • Experience with knowledge distillation, including offline or off-policy distillation, on-policy distillation, self-distillation, sequence-level distillation, logic matching, or representation transfer.
  • Experience designing teacher–student training pipelines or transferring capabilities from large foundation models to smaller models.
  • Experience with multimodal models spanning language, vision, video, audio, or other sensor modalities.
  • Understanding of inference efficiency, memory hierarchy, hardware accelerators, or hardware–software co-design.
  • Strong publication record, influential open-source contributions, or an equivalent record of applied research impact.

Responsibilities

  • Investigate fundamental questions about the design, training, and distillation of multimodal foundation models.
  • Develop and evaluate model architectures, pre-training objectives, data strategies, optimization methods, and teacher–student learning techniques.
  • Explore how to transfer capabilities from large foundation models to smaller, more efficient models.
  • Co-develop frontier and efficient models for Apple silicon and on-device intelligence.
  • Design architectures that distill effectively and study how teacher and student models should be trained together.
  • Develop distillation methods that preserve reasoning, multimodal understanding, instruction following, and other capabilities under constrained model capacity.
  • Incorporate deployment constraints into research from architecture experiments and pre-training through distillation and final model evaluation.
  • Conduct ambitious experiments at scale and translate successful research into foundation-model technologies for Apple products.
  • Potentially contribute to publications and open-source selected models, research artifacts, evaluations, or tools.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service