AI Engineer — LLM Systems

What you'll own

• —Retrieval pipelines end to end: chunking, embedding, ranking, and the evaluation that proves a change helped
• —Model selection on cost, latency and quality — and saying so when an LLM is the wrong tool
• —Guardrails and failure behaviour, including what happens when the model is confidently wrong
• —The inference bill, tracked per client

What we need

• —Python, and LLM systems you've put in front of real users
• —Retrieval done properly — you can explain why your chunking strategy was chosen and how you measured it
• —An evaluation habit: you don't ship a prompt change on vibes
• —Enough backend engineering to own the service around the model

Helpful, not required

• —Fine-tuning and when it beats retrieval
• —Self-hosted open-weight models and the operational cost of them
• —Computer vision or speech alongside text

What this role isn't

So the wrong candidates can rule themselves out in thirty seconds rather than three interviews.

• ×It isn't a research role. There's no publication track here.
• ×It isn't prompt engineering alone — you'll write and operate the surrounding system.

How we interview

1. 01Screening call — 30 minWhat you've deployed and how you knew it worked.
2. 02Systems conversation — 90 minWe design a retrieval system together for a real client-shaped problem, including how you'd evaluate it.
3. 03Conversation with the Head of AI — 45 minJudgement, cost thinking, and your questions.

Decision within two weeks of the first call. If it's a no, you get a reason, not silence.