AI Training Data Engineer

Every model we ship is only as good as its data. Blicker reads utility meters in production every day. Lapwing, our new LiDAR/3D AI product for the construction sector, is learning to understand underground assets. And our client systems in industry and logistics all depend on datasets that were selected, labelled, and monitored with care. Right now that care is spread across our AI engineers, in between their model work. We're looking for the person who makes it their job.

You own the data side of every model we train. Today, our six AI engineers and consultants each carry a piece of it — selecting data, writing annotation instructions, checking label quality, watching for drift. It works, but it doesn't scale, and it isn't the sharpest use of their time. You take it over, run it better than we do now, and build it into something that scales with us.

A normal week looks like: curating the next training set (which of Blicker's new meter types are underrepresented? which Lapwing scans cover the edge cases?), reviewing annotations from our student pool and sharpening the instructions where labels disagree, checking whether production data has drifted since last month and assembling a retraining set if it has, discussing new features for our labelling and data management software with the software engineering team and running the annotation operation itself — scheduling students, onboarding new ones, and resolving the edge cases they surface. You sit in the middle of the AI team, and your work directly determines how good their models can be.

This is a new role. There is no predecessor's playbook — you define how data operations works at Clockworks, with the AI team beside you but the ownership genuinely yours.

• Own the dataset lifecycle for Blicker, Lapwing, and client projects: selection, versioning, quality metrics, dataset management

• Monitor drift in production data and assemble retraining datasets when reality shifts, and build/automate the systems that monitor this

• Write and sharpen annotation instructions, guard label quality, and troubleshoot the edge cases annotation surfaces

• Run and grow our student annotation pool: recruiting from (tech) universities (Delft, Eindhoven, Nijmegen, etc.), onboarding, scheduling, quality feedback

• Build the tooling and process that make all of this repeatable — you decide what good data ops looks like here

You sit in the middle of our AI and software engineering teams. You won't just be managing data; you’ll actively collaborate with the AI engineers to design model experiments based on your data insights, and work closely with software engineers to refine our internal data tools and infrastructure. Your perspective on the data will directly influence the development of our models and the systems that support them.

Our annotation pool is more than labelling capacity — it's also where we find our future computer vision/AI engineers. We want it to become something with a reputation: a bigger pool, a proper communication channel, workplaces in our office for students who want to come in, students joining our Thursday drinks and demos, and a party once/twice a year — for them and for the team. None of this is a requirement of the role; the data work stands on its own. But if building that community sounds like your kind of challenge, it's yours to claim, and we'd love you to claim it.

• Hands-on annotation experience or familiarity with annotation tooling — time in our own pool counts double

• Familiarity with computer vision models and their failure modes

• Organizer energy: communities, events, getting people to actually show up

• Dutch (the student pool is largely Dutch-speaking)

Annotation operations is detail work, and quality problems have a talent for surfacing at the worst possible moment. Students cancel, mislabel, and graduate, the operation has to keep running anyway. Drift monitoring means being the person who spots bad news early and says it out loud. And because the role is new, you'll sometimes be building the road while driving on it. If that energizes you more than it scares you, keep reading.

Clockworks builds AI that simply works. From Het Groothandelsgebouw at Rotterdam Central Station, we have been building AI solutions that run in production since 2017 — no demos that gather dust. From computer vision and classic machine learning to LLMs and agentic AI: we handle the complex machinery under the hood so users get something simple and reliable. We do this in bespoke projects for industry and logistics, and in our own products such as Blicker. We are a team of around fifteen: small enough that your work counts, big enough to build serious systems.

An intro call (30 minutes — mutual fit, role scope, salary confirmed in the open). Then an in-depth conversation with the AI team plus a small, realistic case based on our actual data work — we discuss it together, we don't harvest free work. Then coffee with the founders and a walk through the terms. Offers are written, with your band position and growth path per our published salary framework spelled out. We respond to every application.

Send your CV and a short note on why this role fits (a few sentences is enough — no formal cover letter needed) to careers@clockworks.co. Questions? Contact Victor Westerwoudt directly.

Note: unsolicited services or offers from recruitment agencies will not be responded to.