MK
For larger companies

Specialist engineering for larger companies

One senior engineer for a hard problem in software, data or AI — delivered in fixed-fee phases, each with a written acceptance criterion, and each one you can stop at.

See recent work

Two programmes delivered in 2026

Both are described here as the clients allow. The mechanism is the part worth reading — it is what gets repeated.

Global electronics & healthcare manufacturer · Supply chain / operations planning · 2026

Built a decision-support engine that, per SKU per planning month, recommends how much of a plant's output ships to each regional distribution center to meet demand at lowest total landed cost — as a transparent, auditable linear program a planner can trust and override.

The hard part of decision-support isn't the maths — it's getting planners to trust a new engine over the spreadsheet they've run for years. Instead of asking them to, I built a cross-check that runs the live solver on the exact inputs their spreadsheet used and returns a per-SKU diff of both the demand-split step and the allocation step. The engine could be proven apples-to-apples against the incumbent process, one SKU at a time, on their own numbers — so adoption became a matter of evidence, not faith.

For decision-support that replaces human judgment, being validatable against the old way matters more than being clever.

  • ExplainableEvery recommendation traceable to a named business rule
  • SKU-by-SKUProvable against the incumbent spreadsheet, on the planners' own numbers
  • RepeatableSame inputs, same answer — every run auditable after the fact
  • ExtensibleNew business constraints slot in without rework
Read the full account
Listed satellite-servicing operator · Space / satellite servicing & debris removal · 2026

Built the frontend for a large, CMS-driven corporate and investor-relations platform — editor-composable pages, up to 10 locales, and interactive 3D/WebGL brand experiences.

A single query fetching 15+ section types blows straight past Contentful's 8192-byte query-text and 11,000 cost-unit ceilings — the platform simply refuses it. Rather than cap what editors could compose on a page, I changed the data-access shape: split the fetch into a lightweight page-shell query plus parallel per-section-type batches. Pages could then grow to any number of section types while staying under the platform limits, and every section type became independently cacheable as a bonus.

When the platform's limits fight the product's flexibility, change the data-access shape, not the product.

  • Zero-codeEditors launch and change pages with no engineering involvement
  • 10Languages served — investor content and news across global markets
  • IR-gradeInvestor-relations platform delivered under a strict security posture
  • 3D + gamesInteractive brand experiences without compromising the site
Read the full account

Before this: the payments framework behind four global banks, delivery on a £1Bn government platform carrying 1.3 million criminal cases, and e-commerce at Costco scale through a 200% international expansion.

Start here

Phase Zero

Three weeks · Fixed fee, quoted after the first call

One named person, a fixed fee, and a paper written for the steering group. Where anything can be run, it comes with a proof on your own numbers — read-only, cross-checked against the process you use today.

Review

Something exists and is going wrong: a programme, a vendor implementation, a model. I read the programme artefacts, the vendor's deliverables, the model and the data, and interview the sponsor, the technical owner and two users.

Proof

Nothing exists yet. Take one product family, one plant, or one planning month; build the engine to first light on your data, read-only; run the cross-check against the incumbent spreadsheet line by line; report the diff.

What the steering group receives — a paper, not a deck
  • The decision in one sentence, in the steering group's own words
  • What the evidence says — the diff against the incumbent process, or the state of the programme against its plan
  • Options with a cost and a time against each, including "stop" and "stay with the vendor"
  • A recommendation, and the risks of taking it
  • A costed, phased plan with an acceptance criterion per phase, a named lead, and what your team owns at the end
  • Appendix: method, data used, and a reproducibility statement
It ends cleanly

Phase Zero terminates cleanly. If the paper says do not proceed, or stay with your vendor, that is the deliverable and the engagement ends there.

What follows, if the paper says it should

Two shapes, both delivered as fixed-fee phases. There is no retainer at the end of this — a support-and-extension phase is still a phase, with its own acceptance criteria.

Data & AI engineering

Decision-Support Engineering

An explainable, auditable engine — allocation, planning, pricing, scheduling — that recommends and does not execute. Phase 1 is typically eight to twelve weeks: the engine to production shape on your data, a validation console for your analysts, and no write path into a system of record.

Acceptance criterion

Reproduces your existing process on your own numbers to an agreed tolerance, with every recommendation traceable to a named business rule.

Platform & security engineering

Platform Engineering under a security posture

A large, editor-driven, multi-market, integrated platform — or one scoped platform component inside a systems-integrator programme. Phase 1 is typically eight to sixteen weeks: one platform surface to production, behind your own security review.

Acceptance criterion

Passes your security review and your editor or UAT acceptance, with the tradeoffs documented and handed over.

How it is contracted

The risk in a programme this size is carried by the contract structure, not by a promise. These are the terms, stated before anyone has to ask for them.

  • Fixed fee per phase against a written acceptance criterion. Payment on acceptance, not on the calendar.
  • Acceptance is evidence on your own numbers, or passing your own security review — never a demo.
  • Recommend, don't execute. No write path into a system of record without its own acceptance gate.
  • A named lead. The person who wrote the paper delivers the phases.
  • Termination for convenience at any phase boundary, paying only for accepted phases.
  • A defect warranty on each accepted phase, window and cap set in the statement of work.
  • You own the deliverables. I retain pre-existing tooling — the solver harness, the validation console pattern — which is what lets them arrive already built.
  • Standard short MSA plus a one-page SoW, or your MSA where one exists. Vendor onboarding pack, insurance certificate and a pre-answered security questionnaire ready before the first call. UK and US paper, USD invoicing.

Talk about a programme

A first call establishes the decision in front of the steering group, who signs, and what your discretionary approval limit is — which is what sets the Phase Zero fee. No rate card, and nothing to put out to tender at this size.