Software Inference Deployment Engineer

The Opportunity

Iris servers are going into data centres where the customer already runs a model-serving stack built around frameworks and tooling they are not going to replace. Getting Lumai's software into that stack, and keeping it working when they change something, is now the difference between an evaluation and production. This hire owns the software half of the deployment.

The Role

You will own the software-side integration and customer support of Iris servers in third-party data centre environments. You begin alongside Lumai's software and engineering teams, integrating and hardening the Iris software stack, supporting model onboarding through the toolchain, and getting hands-on with the disaggregated prefill/decode runtime so you understand the platform from the inside. Then you take ownership in the field: supporting customers as they integrate Iris into their own frameworks, and troubleshooting it in production.

What is genuinely hard about this is that you are debugging at the seam between someone else's stack and hardware nobody else runs. The customer's team knows their infrastructure and not ours, you know ours and have to learn theirs quickly, and the fix has to hold when they upgrade their serving stack six weeks later.

By month six, you will have taken models through the Iris toolchain on a customer configuration, integrated Iris into at least one customer's serving stack, and built the troubleshooting path the team reuses in the field.

What You'll Do

• Integrate, test and harden the Iris software stack with Lumai's software and engineering teams ahead of deployment.

• Support model onboarding through the Iris toolchain: loading, format conversion and framework integration.

• Develop hands-on familiarity with the disaggregated prefill/decode runtime and how Iris servers operate alongside decode processors.

• Support customers integrating Iris into their own frameworks and inference workflows.

• Own software-side troubleshooting in the field as the first line of response after deployment.

• Train and enable customer ML and infrastructure engineering teams on the Iris software platform.

• Feed field findings, integration issues and customer feedback back into product and engineering.

What We're Looking For

Must-have

• Hands-on software engineering experience in AI infrastructure, inference serving, accelerator integration or comparable hardware-software environments.

• Strong Python skills and familiarity with major ML frameworks, PyTorch in particular.

• Practical experience with model deployment workflows: loading, format conversion, quantisation or framework integration.

• Comfortable working with inference serving stacks such as vLLM or TensorRT-LLM.

• Familiarity with Linux, containerisation (Docker) and cluster environments.

Strong preference for

• Experience integrating accelerator hardware (GPUs, FPGAs, ASICs, NPUs or novel architectures) into customer inference workflows.

• Familiarity with the NVIDIA inference stack: CUDA, TensorRT, Triton.

• Exposure to disaggregated inference architectures, prefill/decode separation or KV cache management.

• Experience in a customer-facing role, communicating clearly with ML and infrastructure engineering teams.

About Lumai

Lumai is the optical compute company building the next generation of AI infrastructure. Spun out of optics research at the University of Oxford in 2021, we compute with light instead of electrons. Our 3D optical technology carries out the matrix multiplications at the heart of AI inside beams of light travelling through free space, which lets it go beyond the limits of both silicon GPUs and integrated photonics.

In April 2026 we launched Iris Nova, the world's first optical computing system to run billion-parameter large language models in real time, using up to 90% less energy than conventional GPU-based systems. Iris Nova, the first server in the family, is now available for evaluation by hyperscalers, neoclouds, enterprises and research institutions. Aura and Tetra will follow.

Our work won the Falling Walls Award for Science Breakthrough of the Year 2025 and 'Best Overall Technology' at the OCP Future Technologies Symposium. We are headquartered in Oxford.

Why Lumai

You'll work on a new kind of computer. Optical computing for AI has been promised for decades. We have a working system running real models, and the hard part left is taking it to volume.

Your work ships. We are moving from first product to volume production, so what you build this year goes into the servers our customers run.

You'll work across disciplines. Optical engineers, machine learning researchers, and hardware and software engineers solve problems together. You will learn things that don't appear on your job description.

The work matters beyond Lumai. AI's appetite for energy is one of the defining constraints of the next decade. Our mission is sustainable intelligence at global scale: AI that is faster, cheaper to run and far less power-hungry.

You'll join early. You'll have a real say in how we build the product, the team and the way we work.

Equal Opportunity

Lumai is an equal opportunity employer. We make hiring decisions based on skills, experience and potential, and we welcome applications from people of all backgrounds. If you need an adjustment at any stage of the hiring process, let us know and we will do our best to support you.