Principal Performance Modeling Architect

OXMIQ Labs, Inc.; Campbell (CA); On-site; USD 230,000 - 320,000; Full time

Benefits offered by this job

Equity participation

Medical, dental, vision coverage

Competitive base salary

Job summary

OXMIQ Labs, Inc. is seeking a Principal Performance Modeling Architect to own the system-solution performance modeling platform (OxSol) and its advanced analysis capabilities.

This hands-on role drives technical direction, extends the modeling engine to new workloads, and informs architecture decisions and pricing before capital is committed. You will work closely with engineering leadership to deliver defensible, simulation-backed numbers across heterogeneous hardware, influencing both

Qualifications

• 5–8 years of experience in performance modeling, computer architecture, or systems performance engineering.
• Hands-on experience building analytical/first-principles/roofline models of AI or HPC systems.
• Solid understanding of both inference and training workloads and how they stress compute and memory.
• Strong working knowledge of AI accelerator and cluster hardware: GPUs/XPUs, memory hierarchies, and parallelism mappings.
• Python for production-quality code with tests and maintainability.

Responsibilities

• Own and extend the Speed-of-Light modeling platform end to end for performance, energy, and cost-per-token.
• Extend modeling to training workloads and vision-transformer/multimodal models.
• Influence silicon and system architecture by turning designs into defensible, simulation-backed numbers.
• Make the platform extensible and agent-driven, exposing analysis through the OxCapsule interface.
• Validate simulation results against real workload frameworks and tighten model fidelity.

Skills

5–8 years experience

Analytical / first-principles modeling

Inference and training workloads

AI accelerator hardware

Python

Architecture influence & tinkering

AI-assisted development mindset

Strong communication

Education

BS/MS/PhD in CS/CE/EE or equivalent

Tools

vLLM

TensorRT-LLM

Megatron-style parallelism

Job description

OXMIQ designs GPU and AI silicon for large-scale model inference and training, and is building the system software and analysis platforms that prove out those designs before they reach customers. We are a startup driving innovation across the full stack — from atoms to agents — and we move fast on the strength of our own tools.

The Role

The Principal Performance Modeling Architect owns OXMIQ's system-solution performance analysis — the modeling platform (OxSol) and its Speed-of-Light and OxFabric analysis capabilities that convert a proposed silicon and cluster architecture into simulation-backed numbers: performance, energy, and cost-per-token across heterogeneous accelerator topologies. These numbers feed our architecture decisions, our pricing, and our customer proposals directly, and they are produced before we commit capital — which makes this one of the highest‑leverage technical roles at the company.

This is a hands‑on individual‑contributor role. You own the platform's technical direction in partnership with engineering leadership, and you write the hard parts yourself. The platform models LLM inference today; your mandate is to extend its reach to new workload classes and make the engine more extensible as it grows.

We are looking for someone who has seen it and done it — a senior engineer with a tinkering mindset who is energized by being the person who can answer "is this architecture actually optimal, and what does a token cost on it?" and would rather prove it with a defensible model than hand‑wave it. You read simulation output critically, reason about whether an outcome is physically sensible, hold a high bar for code quality, and treat AI‑assisted development as a standard part of how you work.

Key Responsibilities

• Own and extend OXMIQ's system-solution performance modeling platform end to end — Speed-of-Light modeling of performance, energy, and cost-per-token across heterogeneous accelerator hardware and cluster topologies.
• Extend the platform beyond LLM inference to training workloads and vision-transformer / multimodal models, defining the modeling approach for each new workload class.
• Influence silicon and system architecture decisions by turning proposed designs into defensible, simulation-backed numbers ahead of capital commitment.
• Make the platform more extensible and agent‑driven, including exposing analysis capabilities through the OxCapsule interface.
• Validate and reason about simulation results — correlating model outputs against the observed behavior of real workload frameworks (e.g., vLLM, SGLang, training stacks) and continuously tightening model fidelity.
• Set and own the platform's technical direction in line with the executive vision, uphold engineering and code‑quality standards, and provide technical guidance to interns supporting the work.
• Serve as a technical point of contact in selected proposal and partner engagements.

Required Qualifications

• 5–8 years of relevant experience in performance modeling, computer architecture, or systems performance engineering, with a demonstrable, shipped body of work.
• Hands‑on experience building analytical / first‑principles / roofline performance models of AI or HPC systems — deriving models, not only running benchmarks.
• Solid understanding of both inference and training workloads and how they stress compute, memory bandwidth, and interconnect.
• Strong working knowledge of AI accelerator and cluster hardware: GPUs/XPUs, HBM and memory hierarchies, scale‑up/scale‑out fabrics, and how parallelism strategies (tensor/pipeline/expert/data) map onto them.
• Python for building and maintaining a production‑quality codebase — clean abstractions, tests, and the discipline to keep a large platform maintainable.
• A track record of influencing architecture decisions and a tinkering mindset — comfortable getting into datasheets, specs, and code to find out why a number is what it is.
• An AI‑first mindset: fluent use of AI‑assisted development workflows (Claude Code or equivalent), and the judgment to review and reason critically about simulation results, outcomes, and code quality.
• Strong written and verbal communication; able to defend a methodology and explain a surprising result to a non‑specialist.

Preferred Qualifications

• Hands‑on experience operating and modifying modern workload frameworks — vLLM, SGLang, TensorRT‑LLM, or comparable — at the level of understanding their internals and observed performance.
• Experience modeling or optimizing training stacks (e.g., FSDP, Megatron‑style parallelism) and vision-transformer / multimodal architectures.
• Familiarity with LLM serving and efficiency techniques: continuous batching, KV‑cache management, disaggregated prefill/decode, and quantization.
• Background in silicon economics: reading vendor datasheets, die‑yield and cost modeling, and tech‑node tradeoffs.
• Experience with agent / MCP‑based tooling or exposing analysis engines through chat‑native or programmatic interfaces.
• Open‑source contributions to performance, modeling, or inference‑serving projects.
• Experience where a model you built directly informed pricing or capital decisions.

Education

• BS/MS/PhD in Computer Science, Computer Engineering, Electrical Engineering, or a related field — or equivalent practical experience.

This is a hands‑on role reporting directly to the VP of System Architecture and working closely with OXMIQ's silicon, runtime, and systems teams to execute on the technical vision. You will own the technical direction of the platform; you may be supported by one or two interns, but this is not a people‑management role. AI‑assisted development tools (Claude Code or equivalent) are a standard part of engineering practice at OXMIQ and are expected in daily work.

OXMIQ offers a competitive compensation package, including base salary, equity participation, comprehensive medical, dental, and vision coverage, and the opportunity to contribute to foundational silicon and software technology.

OXMIQ is an equal opportunity employer. We evaluate qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, veteran status, or any other legally protected characteristic.