Research Engineer-ML Systems & Compilers (Contractor)
SKILLS
FULL DESCRIPTION
Research Engineer-ML Systems & Compilers (Contractor)
[Employer hidden — sign up to reveal] – [Employer hidden — sign up to reveal] – Cambridge, UK – On-site – Contract
Job Summary
We are a CPU research team working where AI meets low-level systems. We use LLMs, agents and reinforcement learning to optimise CPUs, operating system kernels, compilers and language runtimes on ARM64 platforms. We are looking for an engineer who is comfortable writing a compiler pass, a JIT or interpreter fast path, and a training loop. Ideally you have built at least one of those from scratch because you wanted to understand how it works.
Key Responsibilities
- Design and train RL and LLM-based agents that discover and verify performance optimisations in kernels, language runtimes and compiler output.
- Build the infrastructure around those agents: reward design, sandboxed execution environments, benchmark harnesses, and verification that catches reward hacking.
- Write and optimise performance-critical code, such as compiler passes, interpreter and JIT hot paths, and runtime data structures, and measure it rigorously on real hardware.
- Turn research papers into working prototypes quickly, and turn strong prototypes into patents and publications.
- Work with CPU architects and kernel and compiler engineers to turn ML-driven insights into hardware-software co-design proposals.
Required Qualifications
- BSc or MSc in Computer Science, Electrical Engineering or a related field, or equivalent practical experience.
- Strong C++ and Python.
- Hands-on experience in at least two of the following areas: compiler infrastructure (LLVM, MLIR, or a custom compiler), language runtimes (JIT compilers, interpreters, or dynamic language VMs such as V8, JavaScriptCore, PyPy, LuaJIT or the JVM), training or post-training models, including reinforcement learning (PPO, GRPO, or RLHF-style).
- Evidence of building a non-trivial system end-to-end. Examples include an open-source project, a language, a compiler, a runtime, or a tool you designed yourself.
Please note that visa sponsorship is not available for this contractor role.
Desired Skills
- CPU microarchitecture knowledge: pipelines, branch prediction, caches and prefetching, and how software behaviour maps onto them.
- ARM64/AArch64 and NEON/SVE/SME; simulation with gem5 or QEMU.
- Dynamic language runtime internals: inline caches, hidden classes/shapes, tiered compilation, deoptimisation, garbage collection.
- MLIR dialects and custom lowering passes, or LLVM backend work (instruction selection, register allocation).
- Low-level kernel work in CUDA, Triton or SIMD intrinsics.
- Quantisation and efficient inference (INT4/FP8, GPTQ/AWQ, KV-cache management).
- RL environment design for multi-step, tool-using agents.
- Distributed training (DeepSpeed, Ray, NCCL).
- Publications, patents, or widely used open-source work.
Location
[Employer hidden — sign up to reveal]
About [Employer hidden — sign up to reveal]
[Employer hidden — sign up to reveal] is a leading global provider of ICT infrastructure and smart devices. [Employer hidden — sign up to reveal] has design centers in Cambridge, London, Edinburgh, Ipswich and Bristol. We continue to explore and define new research directions. Apply for this job to join our team.