Skip to main content
TALENT247 // SPECIALIST AI EXECUTIVE SEARCH & TECHNICAL RECRUITMENT
INF · AI Infrastructure

Senior Embedded AI Inference Engineer (CUDA / TensorRT)

London, UK · Full-time · On-site

AI Specialism

AI Infrastructure

Workplace

On-site

Location

London, UK

Compensation

£110k – £150k

annual package

Role Dossier & Specification

REF: SENIOR-EMBEDDED-AI

THE CLIENT & MANDATE

Our client is a premier London embodied-AI company deploying end-to-end multimodal foundation models onto commercial and consumer vehicles worldwide. Running billion-parameter vision-language-action (VLA) transformers deterministically inside tight automotive thermal and millisecond latency envelopes is one of their core engineering differentiators.

They have partnered with Talent247 to appoint a Senior Embedded AI Inference Engineer (CUDA / TensorRT) onsite in London. Sitting between foundation model architects and vehicle platform engineers, the successful candidate will own custom kernel development, low-bit quantization, and deterministic onboard execution across next-generation automotive SoCs.

KEY RESPONSIBILITIES

  • Architect ultra-low-latency C++20, CUDA, and TensorRT / TensorRT-LLM inference pipelines executing multimodal vision-language-action models on automotive-grade NVIDIA Orin and Thor SoCs.
  • Write custom CUDA/PTX and CUTLASS kernels for fused multi-camera patch tokenization, temporal cross-attention, paged KV-caching, and flow-matching action decoding.
  • Co-design INT8, FP8, and INT4/AWQ quantization-aware training and post-training calibration workflows with ML scientists to achieve 3–4x throughput gains with zero safety regression.
  • Eliminate tail-latency jitter through CUDA Graph capture, zero-copy DMA sensor ingestion, deterministic memory arenas, and real-time Linux/RTOS thread scheduling.
  • Build automated hardware-in-the-loop (HIL) profiling suites tracking layer-by-layer roof-line utilization, SRAM/DRAM bandwidth, power draw, and numerical parity against PyTorch reference checkpoints.
  • Support global OEM partner integrations by abstracting hardware-specific graph compilers and safety-island watchdogs behind clean deployment interfaces.

TECHNICAL ENVIRONMENT & OFFERING

Stack: Modern C++20, CUDA, PTX, CUTLASS, TensorRT / TensorRT-LLM, ONNX/MLIR, PyTorch, Nsight Systems & Compute, Linux RT / QNX, automotive GPU SoCs. Workplace: Onsite in London, UK (direct access to vehicle hardware-in-the-loop benches and test fleet). Compensation: £110,000 – £150,000 base salary plus generous equity, private healthcare, and full relocation/visa support.

Core Requirements

ESSENTIAL QUALIFICATIONS

  • 5+ years of systems and inference engineering experience in modern C++ (C++17/20) and CUDA on NVIDIA GPUs.
  • Deep production mastery of TensorRT or TensorRT-LLM, including custom plugin authoring, graph fusion debugging, and mixed-precision calibration (FP16/FP8/INT8/INT4).
  • Demonstrated ability to profile GPU hardware bottlenecks (warp occupancy, shared memory bank conflicts, tensor core utilization) using Nsight Compute and Nsight Systems.
  • Strong understanding of transformer, ViT, and temporal sequence model architectures and KV-cache memory mechanics.
  • Experience building deterministic, safety-conscious real-time software on embedded Linux or RTOS platforms.

PREFERRED / BONUS EXPERIENCE

  • Direct hands-on experience with NVIDIA DRIVE Orin/Thor, Jetson AGX, or similar automotive/robotics edge accelerators.
  • Familiarity with MLIR/TVM/XLA compiler passes, quantization-aware training (QAT) in PyTorch, or speculative decoding on edge hardware.
  • Exposure to ISO 26262 functional safety concepts, zero-copy camera ISP pipelines (GMSL/NvMedia), or robotics middleware.

Position Telemetry

Location London, UK (Onsite)
Department AI Infrastructure
Employment type Full-time
Workplace type On-site
Salary range
£110k – £150k annual package