Senior Distributed LLM Inference Platform Engineer
Bristol, UK · Full-time · Hybrid
AI Infrastructure
Hybrid
Bristol, UK
£95k – £130k
annual package
Role Dossier & Specification
REF: SENIOR-DISTRIBUTEDTHE CLIENT & MANDATE
Our client is a billion-dollar sovereign AI hyperscaler headquartered in the UK, deploying vertically integrated, renewable-powered GPU superclusters across Europe and North America. On top of their bare-metal accelerator fleet, they operate a high-throughput serverless inference and fine-tuning platform serving frontier open-weight language, reasoning, and multimodal models to AI-native startups and regulated European enterprises.
They have retained Talent247 to hire a Senior Distributed LLM Inference Platform Engineer at their Bristol systems engineering hub. Tapping into Bristol's deep parallel-computing and silicon-software heritage, you will architect the multi-node disaggregated serving layer, KV-cache tiering fabric, and intelligent autoscaling router powering trillions of served tokens across their GPU cloud.
KEY RESPONSIBILITIES
- Architect, deploy, and optimize multi-node distributed LLM serving clusters (vLLM, SGLang, TensorRT-LLM, NVIDIA Dynamo) supporting 100B–600B+ parameter MoE and long-context reasoning models.
- Implement prefill/decode disaggregation, RDMA/NVLink KV-cache transfer, distributed prefix-cache routing, and speculative decoding across multi-rack H100/H200/GB200 pods.
- Resolve systemic performance and reliability bottlenecks from GPU kernel stalls and NCCL collective hangs up to Envoy/Rust request scheduling and multi-tenant noisy-neighbor isolation.
- Build reproducible, automated benchmarking harnesses quantifying TTFT, TPOT, goodput under strict SLOs, and tokens-per-megawatt/dollar across newly released open-weight models and hardware generations.
- Evaluate emerging serving engines, quantization kernels (FP8/NVFP4), and RL rollout runtimes, contributing fixes upstream where impactful.
- Collaborate closely with bare-metal Kubernetes infrastructure and product teams to turn raw cluster capacity into effortless developer APIs.
TECHNICAL ENVIRONMENT & OFFERING
Their serving stack combines Python, C++, Rust, Go, PyTorch, vLLM, SGLang, TensorRT-LLM, InfiniBand/RoCEv2 RDMA, Kubernetes, and OpenTelemetry. Working hybrid in Bristol, our client offers £95,000–£130,000 GBP base salary plus meaningful hyperscaler equity, private medical insurance, generous holiday allowance, and flexible human-first working practices.
Core Requirements
ESSENTIAL QUALIFICATIONS
- 5+ years of systems, ML infrastructure, or distributed backend engineering experience operating performance-critical production platforms.
- Direct hands-on experience deploying, profiling, and scaling modern LLM inference engines (vLLM, SGLang, TensorRT-LLM, or Triton) across multi-GPU Kubernetes clusters.
- Strong programming proficiency in Python and at least one systems language (C++, Rust, or Go).
- Deep understanding of LLM serving internals: continuous batching, PagedAttention, KV-cache memory hierarchy, FP8/INT4 quantization, and tensor/expert parallelism.
- Data-driven engineering discipline using Nsight, PyTorch Profiler, Prometheus, and distributed tracing to back architectural tradeoffs with hard numbers.
PREFERRED / BONUS EXPERIENCE
- Experience with disaggregated inference routing (NVIDIA Dynamo, Mooncake, LMCache) or GPUDirect RDMA / Nixl KV-cache transport.
- Upstream contributions to open-source LLM inference engines or kernel libraries.