AI Engineer UK

Worky; Greater London; On-site; GBP 90,000 - 130,000; Full time

Job summary

Nexus Black is building agentic systems to power critical industrial applications. You’ll ship production code from week one and see it used on customer sites, solving real problems in field service and regulation-compliant workflows.

The role emphasizes design, reliability, and end-to-end ownership—from prompts to observability and on-site deployment—within a small, fast-paced team that values practical impact.

Qualifications

• Production experience with LLMs, including system-level prompt design and tool use.
• Established eval pipelines with golden datasets and regression suites.
• Strong fundamentals in Python and TypeScript, Git, testing and CI.
• Ability to work in a small, fast-moving team.
• Curiosity about the physical world and user needs.

Responsibilities

• Design and build agentic workflows using Claude, LangChain/LangGraph and related tools.
• Own evaluation: define "correct" metrics and build evaluation harnesses.
• Move features from prototype to production with observability and tracing.
• Join customer ride-alongs/site visits and apply learnings back to code.
• Collaborate closely with coding agents and maintain docs and PRs.

Skills

Production experience with LLMs

Evaluation habit

Strong software fundamentals

Comfort with ambiguity and pace

Curiosity about the physical world

Tools

Claude Code

Cursor

LangChain

LangGraph

Python

TypeScript

Git

Job description

About Nexus Black

IFS Nexus Black builds AI to solve hard problems for the industries that keep the world turning: up on power lines, down on factory floors, and out in the field. Our products — Resolve (field service and predictive maintenance), Flow (production planning), Comply (regulatory compliance) and Reason (analytical reasoning) — are used by technicians and engineers who cannot afford a tool that doesn't work first time. We pair 40 years of IFS industrial depth with the technical pace of a start-up, and we get results in weeks, not months.

The role

You’ll build the agentic systems behind our products: agents that reason and orchestrate, deterministic code that calculates, and every output traceable. You’ll ship production code from week one — and then watch a field technician use it on a customer site, because that’s where our engineering standard is set.

What you’ll do

• Design and build agentic workflows — multi-step tool use, retrieval, and orchestration — using Claude, LangChain/LangGraph and whatever earns its place next.

• Own evaluation. Define what "correct" means for a workflow, build eval harnesses and golden datasets, and hold the line: if it wouldn’t convince a 30-year veteran, it doesn’t ship.

• Take features from prototype to production: observability, tracing (LangSmith), error handling, and reliability that survives week six, not just the demo.

• Join customer ride-alongs and site visits — plants, hangars, the field — and bring what you learn straight back into the code.

• Work with coding agents daily (Claude Code, Cursor) — they’re how we work, not a novelty — and keep making your own job smaller.

• Contribute like an owner: PR reviews, design docs, and a culture of writing things down.

What you’ll bring

• Production experience with LLMs — you’ve shipped systems using model APIs (Anthropic, OpenAI or similar), not just notebooks: system-level prompt design, agent architectures, tool use, RAG.

• An evaluation habit. You’ve built eval pipelines — golden datasets, regression suites, LLM-as-judge where it’s honest — and you can tell a good demo from a system that holds up in production.

• Strong software fundamentals: Python and/or TypeScript, Git, testing, CI, and the judgement to know when deterministic code beats a model call.

• Comfort with ambiguity and pace — small team, broad remit, shipping over ceremony.

• Curiosity about the physical world. Our users fix aircraft and keep grids running; you should want to understand their job well enough to build for it.

Nice to have

• Retrieval infrastructure at scale (vector stores, hybrid search, reranking).

• Fine-tuning or model-adaptation experience where it genuinely beat prompting.

• Experience in industrial, field-service or other asset-heavy domains.

• Open-source contributions or a portfolio of shipped AI systems.

Why Nexus Black

We’re a small team building AI that has to work the first time, because the people using it are forty feet up a utility pole. AI tooling isn’t a side experiment here — it’s how we work every day. If you want your code in the hands of people who keep the world running, we’d like to talk. Less talk. More impact.