You'll build and operate the systems that turn raw compute into useful intelligence — the inference services that serve LLMs at scale and the data pipelines that feed them. One week you're hunting a tail-latency regression in a production inference service handling millions of requests; the next you're redesigning a Ray Data pipeline so it stops melting down at petabyte scale. The work spans architecture, implementation, and the on-call pager that keeps you honest about both. Researchers and ML engineers will hand you workloads that barely run; you'll hand them back systems that run reliably, efficiently, and cheaply enough to matter.
ResponsibilitiesServe Models at Scale: Design and operate distributed inference systems for LLMs, optimizing throughput, latency, and cost across heterogeneous GPU fleets. Batching, scheduling, KV cache management, autoscaling — you own the levers that make inference economical.
Move the Data: Build large-scale data pipelines (Ray Data, Spark, or equivalents) that ingest, transform, and curate the datasets behind training and evaluation. The bottleneck is rarely where people think it is, and you find it.
Debug the Undebuggable: Chase down the failure modes that only emerge under real production traffic — stragglers, head-of-line blocking, silent data corruption, GPU memory fragmentation — and write the postmortems that prevent the next ten. Define SLOs, build the observability to measure them, and own the on-call rotation that defends them.
Partner Across the Stack: Work directly with researchers and ML engineers to take experimental workloads from "runs on one node" to "runs in production." You're a systems partner, not a ticket queue.
5+ years building and operating distributed systems in production.
Deep experience with at least one large-scale data or compute framework (Ray, Spark, Flink, Beam, Dask).
Strong fluency in Python and at least one systems language (Go, Rust, C++).
Working knowledge of the GPU/accelerator stack: CUDA fundamentals, NCCL, mixed precision, memory layout. You don't need to write kernels, but you should know why a workload is bound by what it's bound by.
Experience operating Kubernetes-based infrastructure, including custom operators or schedulers.
A track record of owning hard production incidents end-to-end — diagnosis, mitigation, and the durable fix.
Bonus: hands-on experience with LLM inference engines (vLLM, SGLang, TensorRT-LLM, TGI), modern lakehouse formats (Iceberg, Delta, Hudi), or open-source contributions to relevant projects.
Above all, we're looking for great teammates who make work feel lighter and aren't afraid to go out on a limb with bold ideas. You don't need to be perfect, but you do need to be adaptable. We encourage you to apply, even if you don't check every box.
About UsMost AI is frozen in place - it doesn't adapt to the world. We think that's backwards. Our mandate is to build efficient intelligence that evolves in real-time. Our vision is AI systems that are flexible, personalized, and accessible to everyone. We believe efficiency is what makes this possible - it's how we expand access and ensure innovation benefits the many, not the few. We believe in talent density: bringing together the best and most driven individuals to push the boundaries of continual adaptation. We're looking for builders and creative thinkers ready to shape the next era of intelligence.
Flexible work: In-person collaboration in the Bay Area, a distributed global-first team, and team offsites.
Adaption Passport: Annual travel stipend to explore a country you've never visited. We're building intelligence that evolves alongside you, so we encourage you to keep expanding your horizons.
Lunch Stipend: Weekly meal allowance for take-out or grocery delivery.
Well-Being: Comprehensive medical benefits and generous paid time off.
