Principal Engineer, Model Performance
- Remote, United States
- Lead
- Posted 10 days ago
or apply from Claude / ChatGPT / Muse
Principal role owning how fast and cheap every model runs on a large GPU fleet, from kernels to serving engines, while building the team around it.
About the role
The fleet spans hundreds of thousands of models and tens of thousands of GPUs across sites in the US and Europe. The measure of good work is new open models served at the best latency and cost within days of release.
At a glance
- Pay
- Not stated in the posting
- Remote scope
- Remote, United States
- Seniority
- Lead
- Skills
- CUDA, TensorRT-LLM, vLLM, SGLang, Quantization, Benchmarking, CI
- Posted
- 6 Nov 2026
- Last checked
- 9 Nov 2026
What you will do
- Own inference performance for the fleet, covering kernels, serving engines, model optimisations and benchmarks
- Set the performance roadmap based on real traffic, cost and upcoming hardware
- Get newly released open models to an optimised production deployment in days
- Choose and shape the serving stack, contributing upstream where it pays
- Run a benchmarking and regression system wired into CI
What they look for
- Experience leading inference performance work at the frontier
- Still hands-on with kernel writing
- Judgement on where the next speedup comes from for a given model and GPU
- Ability to hire, mentor and lead engineers
Nice to have
- Work with open-source engine communities
- Familiarity with TensorRT-LLM, vLLM or SGLang
How Keywordise applies for you
Alex Demo
[email protected]
From your CV
Yes, fully remote
- Sent
- 09:41
- Attached
- Your tailored CV
- 1Upload your CVAdd your CV once. It takes about a minute.
- 2Tailored to this jobKeywordise writes a CV and a cover letter for this job from your real experience.
- 3The form fills itselfName, email, experience and screening questions, field by field.
- 4Application sentYou get a receipt and follow it in your dashboard.
Apply with Keywordise AI7-day free trial, no card.
Apply from your AI assistant
Connect Keywordise to your assistant once, then send it the prompt. It applies for you.
Claude
- In Claude, open Settings, then Connectors, and choose Add custom connector.
- Name it Keywordise and paste the connector URL.
- Click Connect, log in to Keywordise and approve. Then send the prompt in a new chat.
https://keywordise.com/mcpDisconnect any time in your Keywordise settings.
ChatGPT
- In ChatGPT, open Settings, then Connectors. On plans that need it, turn on developer mode first.
- Add a new connector with the connector URL and choose OAuth. Leave the client ID and secret empty.
- Log in to Keywordise and approve. Then send the prompt in a new chat.
https://keywordise.com/mcpDisconnect any time in your Keywordise settings.
Muse
- Open Muse (muse.ai or the app) and send it the message below.
- If Muse asks how to sign in, answer OAuth. Open the sign-in link it sends, log in to Keywordise and approve.
- Send the prompt. Muse asks before it sends anything.
Add a custom connector called Keywordise. It is a remote MCP server at https://keywordise.com/mcp and uses OAuth. Give me the sign-in link.Muse is available in the US only for now.
Apply to Keywordise job q2rc73looaQuestions about this job
Is this job fully remote, and from where?
Yes. The posting lists Principal Engineer, Model Performance as a remote role for candidates in the United States.
What does this job pay?
The posting does not state a salary, so there is no figure to quote.
What experience is needed?
It is a lead role. The posting asks for: experience leading inference performance work at the frontier; still hands-on with kernel writing; judgement on where the next speedup comes from for a given model and GPU. Skills it names include CUDA, TensorRT-LLM, vLLM, SGLang.
How do I apply with Keywordise AI?
Click Apply with Keywordise AI, create an account and upload your CV. Your CV and cover letter are written for this job from what you have really done, then Keywordise applies for you. Trial: 7 days, no card.
More remote engineering jobs
- An AI data and training platform companyDirector of Engineering, Data Labs
- Remote, United States
- NNVIDIASenior Performance Compiler Engineer, Triton
- Remote, United States
- NNVIDIASenior Libraries Engineer, AI and HPC
- Remote, Europe
- NNVIDIASenior Compiler Engineer, AI Inference Platforms
- USD 152,000 to 241,500 per year
- Remote, United States
- NNVIDIASenior Math Libraries Engineer, Direct Sparse Solvers
- Remote, United States
- NNVIDIASenior Math Libraries Engineer - Sparse Linear Algebra
- Remote, United States