Services

MLOps consulting

This page is for technical buyers: heads of data, chief technology officers, and founders who already know what MLOps is and need someone who has actually run it in production. I am an MLOps consultant and machine learning platform architect based in Cape Town, and I have spent a decade building the machinery that keeps models honest after they ship.

What I do

Machine learning operations is the unglamorous half of AI: the training and deployment pipelines, the model registries, the monitoring that notices drift before your customers do. It is also where most AI initiatives quietly fail. A model that performs beautifully in a notebook and cannot be retrained, redeployed or trusted in production is a demo, not an asset.

I design and build that operational layer: ML platforms from first principles, deployment automation, model monitoring, and the architecture reviews that tell you why your current setup keeps breaking. I work hands-on — the person who designs the system writes the code.

The record, in numbers

Ways of working

  • Architecture review. A fixed-fee assessment of an existing ML system or platform: what will hold, what will fall over, and the order in which to fix it.
  • Platform design and build. The full engagement — I design the platform, build it, and stay until your team runs it without me.
  • Fractional AI architect. A few days a month of senior technical leadership: steering an existing team, reviewing designs, and keeping a build honest without a full-time hire.

What I look for in an architecture review

Every struggling ML system I have reviewed fails in one of a few familiar places: training that only one person knows how to run; deployment that takes a week of hand-holding per model; monitoring that watches infrastructure but not predictions, so the model rots invisibly while the dashboards stay green; and a feature layer rebuilt slightly differently for every model, so no two agree on what a customer is. A review names which of these you have, what each is costing you, and the order in which to fix them — smallest intervention first.

The output is written for two audiences at once: an engineering plan your team can execute without me, and a one-page summary your board can read without translation.

Based in Cape Town, working worldwide

I work from Cape Town, South Africa, in a time zone that overlaps the whole European working day. Most of my production work has been delivered remotely for organisations in the United Kingdom and Europe — banks, lenders and retailers whose systems could not be allowed to fail quietly. If you are looking for an MLOps consultant in Cape Town specifically, we can meet in person; if you are anywhere else, the work travels well.

Questions technical buyers ask

What does an MLOps consultant actually do?

MLOps — machine learning operations — is the engineering discipline that takes a model from a data scientist's experiment to a system a business can rely on: automated training and deployment pipelines, monitoring for drift and failure, and the platform underneath it all. As an MLOps consultant I design and build that machinery, and I have run it in production for banks, lenders and retailers for a decade.

Do you only work with companies in Cape Town or South Africa?

No. I am based in Cape Town and work with clients worldwide — most of my production work has been for organisations in the United Kingdom and Europe, delivered remotely with regular overlap hours. Time-zone alignment with Europe makes remote collaboration straightforward.

What does MLOps consulting cost?

Like everything else on this site, the work is scoped and priced as a fixed engagement before it begins — an architecture review, a platform build, or a fractional arrangement of a few days a month. Assessments start from €9,500; larger platform work is priced on scope.

How does this relate to the feasibility assessment?

The feasibility assessment is the front door for new ideas. MLOps consulting usually starts from an existing system: models that take too long to ship, pipelines that break quietly, or a platform that needs designing from first principles. If you are not sure which you need, write to me and I will tell you honestly.

How to start

Write to me about the system you have, or the one you need. The case studies and the writing are public if you want to inspect the record first.

Start a conversation