Inference & Workload Optimization Lead (Horizon AI)

STN • San Francisco Bay Area, CA

1d

About The Position

Horizon AI is building an operating system for organizations, beginning with autonomous data centers. AI agents monitor STN’s GPU infrastructure and take actions when risk is bounded and outcomes are verifiable, while higher-risk decisions are presented as recommendations for a human. Data centers are the initial proving ground for technology intended to generalize beyond infrastructure.

The Inference & Workload Optimization Lead owns the performance of customer AI workloads on committed capacity and builds agent-driven autonomy so optimization expertise can scale across clusters. The role focuses on making customers’ training and inference jobs measurably faster on their dedicated clusters; it is not primarily about serving models to Horizon AI’s own agents.

Customer workload optimization is currently manual and dependent on individual experts. The organization lacks systematic workload profiling, repeatable playbooks, performance baselines, and definitions of good performance by hardware generation. Optimization knowledge is not yet encoded in agents. Internal model serving works but needs maturation, including more cost- and latency-aware routing; GPU capacity for internal inference is negotiated manually against a fleet largely committed to customers.

The role will profile workloads, establish performance baselines beginning with current-generation NVIDIA hardware, encode optimization expertise in agents, work with committed-capacity customers, and take secondary responsibility for internal model serving and GPU allocation advice.

Requirements

• Deep GPU performance engineering experience, including profiling, memory and bandwidth behavior, and kernel-level understanding of performance bottlenecks.
• Serving stack experience, such as vLLM, SGLang, or TensorRT-LLM, including batching, KV cache management, and quantization trade-offs.
• Distributed training and inference experience at scale, including tensor, pipeline, and data parallelism and collective communication tuning such as NCCL.
• Fluency with profiling tools such as Nsight, DCGM, and framework-level profilers.
• Ability to explain performance findings to customers and encode them for an agent.

Nice To Haves

• CUDA kernel work.
• Open-weight model deployment at scale.
• Benchmark design.
• Prior customer-facing technical experience.

Responsibilities

• Profile customer training and inference workloads to identify bottlenecks in areas such as data loading, collective communication, memory bandwidth, kernel efficiency, batching, and parallelism layout.
• Establish performance baselines and define good performance by hardware generation, starting with current-generation NVIDIA hardware.
• Turn optimization expertise into agent-executable optimizations that can be applied across clusters.
• Work with committed-capacity customers, directly or through operating teams, on workload tuning.
• Own internal model serving as a secondary responsibility, including cost- and latency-aware routing and model version migration.
• Advise on GPU allocation between internal inference and customer workloads.

What This Job Offers

• Job Type: Full-time
• Career Level: Principal
• Education Level: No Education Listed