Full-time

Inference Engineer

London, England, United Kingdom

ML Inference Engineer (GPU Performance)

Make in-house models fast and cheap enough to run in production.

Large models are expensive to serve well. When users expect fast responses, latency, throughput and cost per token decide whether the product feels usable.

You'll join an early-stage AI lab that trains its own models and serves them in its own production AI products. Inference engineers here work closely with research, so serving performance feeds into how models are designed, not only how they're deployed.

The opportunity

You'll work on the serving stack behind the lab's models, getting more out of every GPU so users get fast responses at a sensible cost.

This is a hands-on engineering role. You'll go from reading new inference papers to writing kernels and shipping the improvements to production.

Your focus:

  • Build and scale serving pipelines with tight targets on latency and cost.

  • Reduce memory use and raise throughput through quantisation, compression, caching and distributed serving.

  • Write custom GPU kernels for the hot paths, from attention to large matrix multiplications.

  • Work with researchers on architecture choices so models are efficient to serve from the start.

  • Follow new work on attention, KV cache management and batching, decide what's worth adopting and implement it.

What you'll bring

  • An MS or PhD in computer science, machine learning or a related field.

  • Strong production code in Python, C/C++ or Rust.

  • Hands-on GPU programming experience with CUDA, Triton or Metal.

  • Practical experience with quantisation and model compression.

Particularly relevant:

  • Experience with serving frameworks such as vLLM, SGLang, TensorRT-LLM or llama.cpp.

  • CUDA kernel work and multi-GPU communication with NCCL.

  • Experience with inference runtimes such as ONNX Runtime, ExecuTorch or GGML.

Why join?

You'll join at an early stage and help shape the inference stack before it's set in stone.

Research and inference sit close together, so your work affects how the models are built as well as how fast they run.

There's real room to grow, with plenty to learn alongside strong researchers and engineers.

Package & location

Compensation: £150,000-£200,000 + equity
Location: London
Working model: Hybrid, around three days a week in the office

If you've optimised LLM serving or written GPU kernels for production models and want to work close to the research, we'd be interested in speaking with you.

Contact Techire AI for a confidential conversation.

All applicants will receive a response.

Salary

Location

London, England, United Kingdom

Contract

Full-time

Your consultant

Fraser Moran

Full-time

Inference Engineer

London, England, United Kingdom

ML Inference Engineer (GPU Performance)

Make in-house models fast and cheap enough to run in production.

Large models are expensive to serve well. When users expect fast responses, latency, throughput and cost per token decide whether the product feels usable.

You'll join an early-stage AI lab that trains its own models and serves them in its own production AI products. Inference engineers here work closely with research, so serving performance feeds into how models are designed, not only how they're deployed.

The opportunity

You'll work on the serving stack behind the lab's models, getting more out of every GPU so users get fast responses at a sensible cost.

This is a hands-on engineering role. You'll go from reading new inference papers to writing kernels and shipping the improvements to production.

Your focus:

  • Build and scale serving pipelines with tight targets on latency and cost.

  • Reduce memory use and raise throughput through quantisation, compression, caching and distributed serving.

  • Write custom GPU kernels for the hot paths, from attention to large matrix multiplications.

  • Work with researchers on architecture choices so models are efficient to serve from the start.

  • Follow new work on attention, KV cache management and batching, decide what's worth adopting and implement it.

What you'll bring

  • An MS or PhD in computer science, machine learning or a related field.

  • Strong production code in Python, C/C++ or Rust.

  • Hands-on GPU programming experience with CUDA, Triton or Metal.

  • Practical experience with quantisation and model compression.

Particularly relevant:

  • Experience with serving frameworks such as vLLM, SGLang, TensorRT-LLM or llama.cpp.

  • CUDA kernel work and multi-GPU communication with NCCL.

  • Experience with inference runtimes such as ONNX Runtime, ExecuTorch or GGML.

Why join?

You'll join at an early stage and help shape the inference stack before it's set in stone.

Research and inference sit close together, so your work affects how the models are built as well as how fast they run.

There's real room to grow, with plenty to learn alongside strong researchers and engineers.

Package & location

Compensation: £150,000-£200,000 + equity
Location: London
Working model: Hybrid, around three days a week in the office

If you've optimised LLM serving or written GPU kernels for production models and want to work close to the research, we'd be interested in speaking with you.

Contact Techire AI for a confidential conversation.

All applicants will receive a response.

Your consultant

Fraser Moran

Full-time

Inference Engineer

London, England, United Kingdom

ML Inference Engineer (GPU Performance)

Make in-house models fast and cheap enough to run in production.

Large models are expensive to serve well. When users expect fast responses, latency, throughput and cost per token decide whether the product feels usable.

You'll join an early-stage AI lab that trains its own models and serves them in its own production AI products. Inference engineers here work closely with research, so serving performance feeds into how models are designed, not only how they're deployed.

The opportunity

You'll work on the serving stack behind the lab's models, getting more out of every GPU so users get fast responses at a sensible cost.

This is a hands-on engineering role. You'll go from reading new inference papers to writing kernels and shipping the improvements to production.

Your focus:

  • Build and scale serving pipelines with tight targets on latency and cost.

  • Reduce memory use and raise throughput through quantisation, compression, caching and distributed serving.

  • Write custom GPU kernels for the hot paths, from attention to large matrix multiplications.

  • Work with researchers on architecture choices so models are efficient to serve from the start.

  • Follow new work on attention, KV cache management and batching, decide what's worth adopting and implement it.

What you'll bring

  • An MS or PhD in computer science, machine learning or a related field.

  • Strong production code in Python, C/C++ or Rust.

  • Hands-on GPU programming experience with CUDA, Triton or Metal.

  • Practical experience with quantisation and model compression.

Particularly relevant:

  • Experience with serving frameworks such as vLLM, SGLang, TensorRT-LLM or llama.cpp.

  • CUDA kernel work and multi-GPU communication with NCCL.

  • Experience with inference runtimes such as ONNX Runtime, ExecuTorch or GGML.

Why join?

You'll join at an early stage and help shape the inference stack before it's set in stone.

Research and inference sit close together, so your work affects how the models are built as well as how fast they run.

There's real room to grow, with plenty to learn alongside strong researchers and engineers.

Package & location

Compensation: £150,000-£200,000 + equity
Location: London
Working model: Hybrid, around three days a week in the office

If you've optimised LLM serving or written GPU kernels for production models and want to work close to the research, we'd be interested in speaking with you.

Contact Techire AI for a confidential conversation.

All applicants will receive a response.

Your consultant

Fraser Moran

techire AI website footer image