Expected to openData science

Machine Learning Engineer

Classification and detection models for compliance and financial crime, taken from research through to something that survives production traffic.

About the role

This seat opens as Agentic Data Compliance and Korva move from research into production. We are talking to people now so the search is short when it does.

The interesting part of this role is not model accuracy in a notebook. It is making a model's output defensible to someone who has to act on it, and keeping it honest once real data starts arriving.

Responsibilities

  • Take classification and detection problems from framing through to a deployed, monitored model.
  • Build the evaluation harness before the model, and keep it trustworthy as the data shifts.
  • Work with the architect on how models are served, versioned and rolled back.
  • Make model behaviour explainable to compliance and investigation users who will not accept a black box.
  • Watch for drift, bias and the failure modes that only appear at production volume.

Requirements

If you meet most of these and the rest looks learnable, apply. We would rather read your application than have you rule yourself out.

  • Production machine learning experience, not only research or coursework.
  • Strong Python and the surrounding ecosystem.
  • Solid grounding in evaluation: what to measure, and what a metric hides.
  • Experience deploying and monitoring models rather than handing them to someone else.
  • The judgement to know when a simpler method is the right answer.

Nice to have

  • Work on fraud, AML or anomaly detection.
  • Experience with LLM evaluation and retrieval systems.
  • Familiarity with regulated data handling and the constraints it puts on training data.
  • Comfort explaining a model to a non-technical audience.

How you would work here

01

Craft is the point

Work is judged on whether it holds up, not on how it was presented in a status meeting. If you do careful work it will be visible here in a way it never is at scale.

02

Real ownership

You own a product surface rather than a slice of a backlog. That means the decisions are yours to make, and the consequences are yours to sit with.

03

New technology, clear guardrails

There is genuine freedom to try things, set inside limits drawn by people who have taken systems into production before. Experiment, but know where the edges are.

04

Hybrid in Amman

Three days in the office and two from home. The office days are for the work that goes better with everyone in a room, which is most of the hard parts.

How we hire

Five steps, about two weeks end to end. The work assessment is a real day of work and it is paid.

  1. CV review

    We read it properly. You hear back either way.

  2. Call with Essa

    Forty-five minutes with the founder. Your work, our work, and whether the fit is real.

  3. Paid work assessment

    One day on a realistic problem, paid at a fair rate. The closest either of us gets to trying the job.

  4. Team session

    Time with the people you would work beside, so the decision runs in both directions.

  5. Offer

    Made quickly once we are sure. About two weeks from first contact to here.

About Digital Composition Lab

DCL is a founder-led R&D software product development lab. We take on problems where software has to be trusted rather than merely shipped, and we build the products ourselves: Agentic Data Compliance for UK GDPR alignment, and Korva for financial crime prevention.

The lab is small and deliberately full-stack in the widest sense: research, product, design and engineering in one team. More about who we are.