Lumai develops PCIe-form-factor optical processors that perform core AI arithmetic in free-space 3D optical beams using low-cost components to enable high-performance, energy-efficient LLM inference.
THE OPPORTUNITY Lumai is redefining how the world computes. We are an ambitious, venture-backed UK startup pioneering a breakthrough AI accelerator for data centers which uses 3D optical compute. Our radical technology uses light to perform computation at orders of magnitude faster speeds and at far greater scales than ever before, all whilst consuming far less energy than traditional approaches. Lumai is unlocking performance and efficiency gains that could transform the economics of AI and compute infrastructure and reshape how intelligence scales globally. If you are passionate about bringing groundbreaking technology to market, and want to be part of a team pushing the boundaries of what is physically possible, Lumai is where you can make it happen. ABOUT LUMAI Founded in 2022, Lumai is a University of Oxford spinout using optical processing to accelerate large language models (LLMs) and other transformer-based AI systems. The team combines expertise in optical computing, machine learning, and physics. Lumai has already secured over $15 million in investment from leading deep-tech investors like Constructor Capital, IP Group, PhotonVentures and government grants, and is scaling rapidly to deploy the fastest optical compute currently available globally. THE ROLE We are building custom AI hardware and the full-stack software ecosystem to run it. As our first dedicated MLOps Engineer, you will own the infrastructure that takes models from research to silicon-validated production — designing, building, and operating the pipelines, tooling, and platforms that let our AI and hardware teams move fast without breaking things. This is a high-impact, high-ownership role at the intersection of ML research, compiler stacks, and novel hardware. WHAT YOU'LL DO - Design and operate end-to-end ML pipelines: data ingest, training, evaluation, quantisation, and deployment onto custom AI accelerator hardware - Build and maintain experiment tracking, model registry, and versioning infrastructure (e.g. MLflow, W&B, or equivalent) tuned to our hardware-in-the-loop workflows - Own CI/CD for ML: automated testing of model correctness, numerical accuracy, and on-chip performance after every change to models, compilers, or firmware - Develop and maintain tooling for benchmarking model inference on custom silicon, including latency, throughput, power, and utilisation metrics - Collaborate closely with ML researchers, compiler engineers, and hardware architects to identify and remove bottlenecks across the model-to-chip workflow - Instrument and monitor production inference deployments; design alerting and rollback strategies appropriate to hardware-accelerated serving - Manage compute resource scheduling across on-premises accelerator clusters and cloud (GPU/CPU) for training and simulation workloads - Drive infrastructure-as-code practices: containerisation, orchestration (Kubernetes/Slurm), and reproducible environment management - Contribute to the internal developer platform: self-service tooling, documentation, and runbooks that raise engineering productivity across the company WHAT WE'RE LOOKING FOR Must-Have - 5+ years of software or infrastructure engineering experience, with at least 2 years in an ML or AI-adjacent role - Strong Python skills and familiarity with major ML frameworks (PyTorch or JAX); comfortable reading and modifying model code - Hands-on experience building and operating ML pipelines in production: data pipelines, training orchestration, evaluation, and serving - Experience with experiment tracking and model lifecycle management tools (MLflow, W&B, DVC, or similar) - Solid understanding of containerisation (Docker) and orchestration (Kubernetes or Slurm) for distributed compute workloads - Infrastructure-as-code mindset: Terraform, Ansible, or equivalent; CI/CD pipelines (GitHub Actions, Jenkins, or similar) - Experience with hardware-accelerated compute (CUDA/GPU workflows, profiling, performance tuning) — even if not on custom silicon - Strong debugging and observability skills: distributed tracing, logging, metrics dashboards - Ability to work effectively in a fast-moving, ambiguous environment where the hardware and software are both being built simultaneously Strong Preference For - Experience with custom or novel accelerator hardware (FPGAs, ASICs, NPUs, or research chips) - Familiarity with ML compiler stacks: MLIR, LLVM, TVM, XLA, or vendor-specific compilers (NVCC, TensorRT, etc.) - Experience with model optimisation techniques: quantisation (INT8/INT4/FP8), pruning, distillation, or mixed-precision training - Background in on-chip performance profiling and roofline analysis - Exposure to chip bring-up workflows: running early software stacks on pre-silicon simulation or first-silicon hardware - Contributions to open-source ML infrastructure or compiler tooling - Experience in a deeptech, semiconductor, or hardware startup environment COMPENSATION & BENEFITS - Highly Competitive Salary: We are not saying our salary is a blank check, but let's just say it won't be a source of your stress - Share Option Scheme: We are all in this together! We believe in shared success while we build the Lumai of tomorrow - Pension Scheme: Plan for retirement with AVIVA - Private Health Insurance: We firmly believe that you come first, and a happy you is a healthy you! Look after yourself and your loved ones with AXA - Cycle to Work: Spread the cost of a bike, a bike and accessories or just accessories and save on tax - L&D Allowance: Stay at the forefront of your field with a £500 annual development budget - Subsidised On-site Lunches: Enjoy on-site healthy meals at half the price, as Lumai covers 50% of the cost - Holidays: Enjoy some deserved "me time" with 25 days paid holiday (plus bank holidays) per year - Socials: Be part of an inclusive community enjoying occasional all-company off-si