Generative AI Research Engineer, Reasoning u0026 Memory - SIML

Position: Staff Generative AI Research Engineer, Reasoning u0026 Memory - SIML
The System Intelligence and Machine Learning (SIML) Content Understanding teams are seeking a Staff Applied Researcher in Reasoning u0026 Memory Systems. You will be working alonside teams that are in charge of operating system wide embeddings, personalized RAG work streams, tool calling, context compaction / efficiency u0026 memory systems. Projects are focussed on advancing Apple Intelligence capabilities, while working closely across disciplines with our partners in hardware engineering, design and product.

Selected references to our prior work (a) (Use the "Apply for this Job" box below). (b)  (c)

We are seeking a candidate with a track record in algorithm development for agentic reasoning u0026 memory. Key attributes expected in the role are fluency in algorithm development (prompt optimization and post training), strong expertise with relevant techniques (reinforcement learning, multimodal reasoning), and experience with automatic evaluation approaches for agentic workflow. The role includes the opportunity to partner with world class system engineers to prototype and incorporate bleeding edge algorithmic innovations in the context of emerging agentic experiences.

Ability to interface with large scale modeling u0026 data infrastructure is desired.

Experience with building u0026 deploying Multimodal-LLMs Familiarity with distributed training and large-scale data infrastructure

PhD, or MSc in Computer Science/Electrical Engineering, or a related field (mathematics, physics or computer engineering); with a focus on machine learning, or comparable professional experience Strong ML and Generative Modeling fundamentals Strong expertise in one of the following:
Reinforcement Learning, Multimodal Training, Pre-training / Post-training foundation models Proficiency in using ML toolkits, e.g., PyTorch Proven track record of research contributions demonstrated through publications in top-tier conferences, or open source contributions to algorithm