We are looking for a product-minded AI engineer to turn what our customers are trying to decide into agents they actually rely on. AskWaive, our in-product assistant, is already live; the work now is quality, coverage, and an experience that wows.
This is not a role for a prompt author or a demo builder. We want an owner: someone who takes a customer need—already expressed through our serving layers and semantic layer—and turns it into an endpoint, a tool, and an experience that makes the customer measurably better at their job. Agent engineering, deployment, and testing are all part of delivery, not a handoff.
Our core data platform is owned by the team that built it. You'll build on top of it—and where data or infrastructure is holding an agent back, you'll have the standing to propose the change and drive it.
What You'll Do
Own the Outcome: Start from the customer, and stay with it until the agent is answering their needs better
Build Agents and Tools: Build production agents and the tools they call—retrieval, multi-step workflows, action-taking—on Amazon Bedrock, with Bedrock AgentCore as our runtime and tools exposed over MCP (Model Context Protocol).
Turn Serving Layers Into Capability: Our semantic layer and service APIs are where customer needs are already expressed. Your job is turning them into endpoints and tools an agent can wield well—and knowing which metrics and entities it should reach at all.
Shape the Experience, Not Just the Endpoint: How an answer arrives—its speed, its confidence, its explanation, the next step it offers—is the product. A technically correct response a customer ignores is a failure.
Measure Groundedness: Build the eval harness—labeled test sets, groundedness and hallucination rates you can put a number on, and CI gates that stop a regression before it ships.
Measure What Customers Got: Adoption, feedback, intent—then report realized impact back to Product and the business in numbers that hold up.
Diagnose and Iterate: Instrument the tracing that shows why an answer was wrong, turn each failure mode into a specific fix, and ship again. This loop is the job, not a phase of it.
Ship and Operate: Own the serving path—MCP endpoints, latency, cost per response, caching, and runtime guardrails. Deployment and testing are part of the job, not a handoff.
Embed and Influence: Partner with Product Managers and engineers on which decisions an agent should support—and which it shouldn't. Where data or infrastructure limits what an agent can do, drive the change alongside the team that owns the platform.
From Customer Need to Product Superpower
Prove It and Make It Felt
Deliver
Who You Are
The Owner: You measure your work by whether customers made better decisions with the agent's help, not by whether it shipped.
The Agent Builder: You've taken an agent past the demo. You know the hard part is everything after it works once—coverage, grounding, cost, and the failure modes you only see at scale. You fix problems without waiting for a ticket.
The Experience Obsessive: You care how an answer lands, not just whether it was right. Trustworthy, fast, and legible beats clever.
We know that great talent comes from many backgrounds. If you are a builder who cares about the "why" behind the code, we want to hear from you!
HOW WE WORK
We build with coding agents. You set direction and targets, review output critically, and build the harnesses—scaffolding, context, tests, review loops—that make the next task faster for the whole team. The leverage is in the verification: the tests, CI gates, and typed interfaces that make agent-written code safe to merge.
What You"ll Bring
Experience: 5+ years building production software or data-intensive systems used by real customers, with strong Python skills—including at least one LLM or agent feature you shipped to production users.
AWS & Agent Stack: Hands-on AWS experience, ideally Amazon Bedrock and Bedrock AgentCore. Equivalent agent runtimes transfer, but we build on AWS and you'll be productive faster if you have prior exposure.
Agent Engineering: Orchestration, tool and function design, context assembly, and multi-step workflows—you've made the design calls, not just the prompt calls.
Retrieval & Domain Grounding: You've built retrieval over real business data, you can explain how context gets selected and why an answer was wrong, and you have enough SQL and metric fluency to catch an agent answering with the wrong number.
Evaluation & Iteration: You've built or owned an eval loop for a generative system—labelled sets, regression gates, and a number you could defend when someone asked whether it got better.
Product Sense for AI: You have opinions about how an agent should behave—when to answer, when to ask, when to refuse, how to show its work—and you've shipped those opinions, not just argued them.
Measurement & Narrative: You've quantified what an agent actually delivered for users—adoption, task success, decision quality—and presented it to people who were never going to read your traces.
Working With Agents: You've used coding agents on real production work, you can tell correct output from merely plausible output, and you invest in the harnesses that make the next task faster.
Communication & Collaboration: Excellent communication and collaboration skills, with a track record of working across technical and non-technical teams.
LLM observability tooling (Langfuse, LangSmith, or Arize).
Semantic layer or metrics-layer design (Ossie, dbt).
Field-service, logistics, or other operational SaaS domains.
Prior experience as a senior or lead engineer responsible for guiding technical direction.
Bonus Points For:
Location: Remote (UK-based) with occasional collaborative team meetups.
Right to Work: Applicants must currently have the legal right to work in the UK. We are unable to provide visa sponsorship for this position at this time.
