Staff/Principal ML Infrastructure Engineer

Quick Facts

• Role: Staff/Principal ML Infrastructure Engineer
• Focus: AI-native table format, indexing, transactional metadata layer, and high-performance ML training data loading

Description

As a Staff/Principal ML Infrastructure Engineer, you will architect the on-disk format, indexing structures, transactional metadata layer, and a Rust data plane plus training data loader used for Apple’s largest ML training workloads. You will build AI-native table and indexing primitives that support efficient global shuffling and low-latency random access across multi-petabyte datasets. You’ll also integrate storage and table contracts with JAX (Grain), PyTorch, and Hugging Face for multiple training stages.

Responsibilities

• Design the next-generation AI-native table and indexing format: on-disk layout, snapshot isolation and time-travel metadata, optimistic concurrency control (OCC) for concurrent writers, zero-copy selection/deletion bitmaps, and multimodal column-group evolution
• Build compressed global indexing and random-access primitives for global shuffling and single-RTT ranged object-store reads across tens of millions of files (ArrayRecord, Parquet, Lance)
• Lead the core Rust data plane and high-performance training data loader, including off-GIL async transport, zero-copy Arrow and record decoding, range coalescing, and tail-latency request hedging
• Integrate storage engine internals with ML training frameworks by owning storage-to-table layer contracts for JAX (Grain), PyTorch, and Hugging Face workloads (pre-training, mid-training, SFT, RL)
• Validate system-level designs with hardware and network benchmarking; prove throughput, memory footprint, QPS scaling, and tail latency on real cluster workloads
• Raise engineering bar by mentoring senior engineers and leading Rust/Python design and code reviews; partner with research and data-platform teams

Requirements

• 10+ years of experience building storage engines, table/file formats, database internals, or high-performance distributed data and ML systems
• Demonstrated experience designing and implementing table or file format internals (e.g., Iceberg, Lance, Delta Lake, Parquet, Arrow, Vortex, Riegeli/ArrayRecord, RocksDB/LSM trees, or custom engines)
• Strong production systems programming in Rust or C++ (Rust strongly preferred), plus Python/PyO3 for ML framework bindings
• Deep hands-on expertise in low-level I/O and memory performance engineering: mmap, zero-copy memory layouts, SIMD/vectorized decoding, RoaringBitmaps, and async runtimes (tokio)
• Experience with high-throughput S3 and GCS object-store client internals (range coalescing, connection pooling, tail-latency hedging)
• Strong understanding of transactional storage semantics: snapshot isolation, MVCC, optimistic concurrency control, manifest trees, and metadata scaling
• Working knowledge of large-scale distributed training data access patterns across GPU/TPU meshes (global shuffling, sequence packing, deterministic checkpoint/resume)
• Track record of setting technical direction across teams and shipping production systems at petabyte scale
• B.S., M.S., or Ph.D. in Computer Science/Computer Engineering (or equivalent practical experience)

Benefits

Not specified in the provided text.