Reducto Machine Learning Infrastructure Tech Lead at Reducto owning systems for high-performance model training and inference. Lead infrastructure projects and set technical direction while hands-on building and optimizing infrastructure.
Responsibilities
As our ML Infrastructure Tech Lead, you'll own the systems that make high-performance model training and inference possible at Reducto.
This is a deeply hands-on role: roughly 80% of your time will be spent building, debugging, and optimizing our infrastructure. The remaining 20% will focus on setting technical direction - identifying bottlenecks, planning our infrastructure roadmap, and helping the ML and Platform teams make strong architectural decisions.
You'll work across the stack, from model-serving kernels and GPU utilization to distributed systems and Kubernetes. We're looking for someone with the experience and judgment to lead ambiguous, high-impact infrastructure projects while remaining close to the code.
This is a fully in-person role at our San Francisco office.
Own the technical direction and roadmap for Reducto's ML infrastructure.
Build and maintain our training and inference stack, balancing fast experimentation with high-performance production serving.
Optimize model serving at every layer, including kernels, runtimes, batching, scheduling, and distributed inference.
Design systems for reliable multi-node, multi-GPU training and inference.
Improve GPU utilization, latency, throughput, reliability, observability, and cost efficiency.
Develop benchmarks that identify bottlenecks and guide infrastructure investments.
Evaluate state-of-the-art advances in training and inference and apply the ones that matter.
Build the tooling and abstractions that help ML engineers move quickly from experiments to production.
Qualification
Have optimized or implemented CUDAHave built observabilityImpactSpeed
Required
5+ years of experience building production infrastructure, including significant ML systems experience
Have led complex technical projects from an ambiguous problem through production deployment
Strong Python and systems-engineering skills
Understand the performance characteristics of modern GPU training or inference workloads
Comfortable with Kubernetes and distributed training or serving
Preferred
Have optimized or implemented CUDA, Triton, or custom model-serving kernels.
Have contributed meaningfully to frameworks such as vLLM, SGLang, PyTorch, TensorRT-LLM, Ray, or related open-source systems.
Have operated distributed inference or training across hundreds or thousands of GPUs.
Have built observability, scheduling, or capacity-management systems for GPU workloads.
Have experience at an early-stage or high-growth startup.
Care deeply about connecting technical excellence to measurable business impact.
Impact: Your work directly shapes how the world’s best AI companies access and use enterprise data.
Speed: We move fast, ship often, and iterate in days, not months.
