Head of Inference, Perimeter Compute

Head of Inference

Full Time, Remote, NYC Preferred (US Based)

About Montauk Capital

Montauk Capital builds and backs companies at the forefront of the Electron Economy, the generational shift towards electrified, intelligent technologies reshaping industries and driving unprecedented demand for energy. Our team combines deep investing acumen with decades of operating experience to give founders the strategic clarity and hands-on support that accelerates the building of enduring companies of consequence.

About Perimeter Compute

Co-founded by Montauk Capital, Perimeter Compute is a pre-seed venture specialized in modular, metro-edge AI capabilities. By leveraging existing infrastructure for inference deployment, Perimeter Compute provides low-latency, SLA-guaranteed performance across diverse GPU SKUs and colocation environments. Our technology intelligently routes traffic based on demand proximity and real-world network limitations, bypassing the heavy power and infrastructure requirements of traditional hyperscalers. Currently initiating operations with pilot nodes in NYC, we are executing a city-by-city expansion strategy with plans for a broader multi-metro rollout.

About the Role

We are seeking a visionary and execution-oriented Head of Inference. You'll define the inference architecture, make foundational decisions, build the first POC, and own this domain end to end alongside the CEO. You will be a senior, hands-on technical leader and the technical authority on inference in the room. You’ll own the key technical decisions, and will be the internal and external expert on inference. You will own the core inference capability driving the platform and customer experience, and have a strong voice over the technical foundation of the company. You’ll evolve the vision into a viable proof of concept, building the practical system to then design and implement distributed inference systems. Alongside the CEO, you’ll represent the company with top-tier partners, early customers and investors, and will own this domain end to end. In addition to the CEO, you will have the support of a team of strong advisors, and the initial founding team.

What You’ll Do

Create the inference strategy and define the inference architecture for Edge AI

Own the inference serving layer end-to-end: vLLM, TensorRT-LLM, Triton, or equivalent

Build a credible POC fast — proves the platform works to NVIDIA, cloud providers, and customers

Drive cost-per-token optimization

Optimize GPU utilization, KV-cache management, and batching for production workloads

Own observability and reliability SLAs

Build distributed inference pipelines across multi-GPU, multi-node edge deployments

Set performance baselines and SLAs for inference latency and throughput, plus observability and performance SLA’s

Define quantization strategy

Translate complex inference requirements for infrastructure designs

Define the software access layer architecture and oversee integration efforts

Engage credibly with investors, partners, and technical stakeholders, represent the company externally

What You’ll Bring

You have a passion for inference and a background as a hands-on technical builder who has directly implemented inference systems before, ideally in production or near-production environments. Deep knowledge and are excited about model serving, and the practical engineering required to make an inference system work on real hardware. You can take a vision and initial concept and translate it into a viable POC quickly and are comfortable making foundational technical decisions quickly, in ambiguity, and building first of a kind.

If inference is your craft and you've built systems in production, we want to talk.

Production inference serving — vLLM, TensorRT-LLM, Triton Inference Server, or equivalent distributed at scale

Quantization, SGLang, containerization, cost-per-token

Observability tooling:distributed tracing, latency profiling, alerting. Instrument and debug complex distributed systems with a focus on building world-class observability and debuggability tools

C++/CUDA/Rust

GPU utilization and CUDA kernel optimization — has pushed hardware to its limits

Batching, KV-cache, speculative decoding expertise

Scale systems using Kubernetes, Ray, custom load balancing, multi-GPU/multi-node inference

Has built a serving system that NVIDIA and cloud providers respect

Model deployment and serving

Systems engineering

Technical leadership experience, either over teams or outcomes

Startup / 0→1 DNA: You ship fast and communicate clearly

Why Join Us

Category-Defining Opportunity: Solving the AI inference bottleneck without the burden of power and infrastructure constraints Own the metro edge inference across heterogeneous, disparate compute nodes

Massive Market Opportunity: AI spending projected to exceed hundreds of billions annually, 54GW of AI Inference demand expected by 2030

Studio Support: Leverage Montauk Capital's resources, network, and operational expertise during critical early stages

Competitive compensation + equity: True ownership over what you build