Platform engineer

The work

• Own the knowledge-base and retrieval layer: document pipelines, embeddings, and the record of what a firm knows and how sure we are of it.
• Build context connectors for CRMs, drives, notes and transcripts that respect the client's permissions to the letter.
• Design the evaluation harnesses that decide whether a claim carries its source or gets held back. When the morning brief says "3 items held back: low confidence", that is your code deciding.
• Keep us model-agnostic: routing, benchmarks, and the discipline to swap a provider the quarter it stops being best.
• Care about latency and cost like someone who has read the invoice.

You, probably

• Strong Python and real production data systems: Postgres at minimum, plus pipelines that survived their second year.
• You've built retrieval and evaluation for LLM systems and can say precisely where yours lied to you.
• Opinions about provenance. If a number has no source, you want the system to say so, loudly.
• Comfortable being the second reader on deployed work and the first author on the platform.
• Allergic to demos that can't ship.

Pay and terms

• £100,000 to £140,000 depending on what you've built, plus meaningful equity.
• London. Hybrid, with more desk and fewer site visits than the deployed role, by design.

One limit, volunteered

Much of this work is invisible when it's right; nobody screenshots the retrieval layer. If you need the demo to be yours, the forward-deployed role will suit you better. We hire for both.