Forward Deployed Engineer

Half your time is with customers and half is building. You'll sit with researchers, analysts and ops teams, see where their week disappears, and find the problem that matters most. Then you figure out what good looks like, build it, ship it, and see how it performs.

You own the work from the first conversation to production. It's like running a company inside a company, with highly experienced founders sitting next to you.

• —A research system used by one of the world's largest NGOs. It writes and runs its own analysis code over millions of documents, figures out the salience of events and links its claims back to their sources.
• —Data analysis systems for making sense of long running survey data with natural language and dynamic AI powered UIs.
• —Our own models for estimating how whole populations would answer survey questions nobody has asked yet.
• —Sargasso, our agentic engineering platform. Agents build alongside us, and everything they ship passes the same checks a person's work would.

We decide what “good” means before we build anything. Every bug or user correction becomes a permanent test. We publish our misses alongside our wins. Every result needs an error bar, and when something fails, we want to hear why.

• —Shipped production software you owned end to end, ideally in Python and TypeScript.
• —Built real things with LLMs: agents, retrieval, evals, fine-tuning, or open-weight models running well.
• —Developed the instinct to write the eval before the feature, and you know exactly what a held-out set is for.
• —Started something, whether a company, a product, an open-source project or a side hustle, where the outcome was yours.
• —Explained a technical trade-off to someone non-technical and had them actually get it.
• —Been a power user of LLMs / AI for a long time.
• —Developed a reputation as someone with a bias for action, someone who gets things done.

If you're strong on most of this, apply.

Please only put things in your application that you can talk about confidently without checking with an LLM first. We are happy for you to use an LLM in the interview but fair warning, if you act like a meat proxy, we will not hire you.

We're small, tight-knit and highly autonomous. Founders Nick Fellingham and Andy Littledale have built together for fifteen years: SecondSync (acquired by Twitter), Twitter itself, broadcast AI with the BBC, and now Elver.

Stack: Python, FastAPI, Postgres/pgvector, Next.js, TypeScript, frontier APIs, open-weight models (Llama, Gemma, Qwen), LoRA, vLLM. We pick tools per problem.

Package: around £120k, plus benefits. You'll be in Bristol with the team 1-2 days a week, and with customers when the work needs it.