Engineering Director · Agile AI Practitioner

Turn a chaotic R&D team into predictable delivery.

“Most AI projects die in the demo. I take them the last, hardest mile into production: monitored, documented, and honestly reported. No watermelons.”

Book a 20-min call → Explore services →

01 · Services

Four ways I take AI the last mile.

Interim Engineering Director

I run your engineering function until it runs itself.

Who it's for

Organizations where capable teams keep missing dates because no single person owns shipping end to end. The talent isn't the problem. Accountability is.

What you get

Install velocity, capacity and burn-rate tracking from week one
Lead cross-functional delivery and shield engineers from organizational noise
Enforce an Industrial Definition of Done across every workstream

Result promise

Within 90 days, your delivery dates are commitments, not guesses.

Agile AI Transformation

I move your AI from notebook to production with ROI you can measure.

Who it's for

Teams stuck in PoC purgatory — endless experiments, no production deployment, burning R&D budget.

What you get

Reframe sprints to validate hypotheses, not ship features (Agile for probabilistic systems)
Build the MLOps and Quality Gates that block non-compliant releases
Ship one prioritized model to production with monitoring and documentation

Result promise

First model live in production within 90–120 days.

Watermelon Check

An independent read on whether your AI project is really green, or red under a thin green skin.

Who it's for

Boards and investors told a project is “on track” who need verification before the next funding tranche or go-live.

What you get

Interrogate velocity, burn-rate and Definition of Done against reported status
Pressure-test the architecture and delivery claims directly with the team
Deliver a one-page true status with the specific risks being hidden

Result promise

Real project status in 5 working days, in writing.

Vendor Due Diligence

I write the requirements vendors can't wriggle out of, then tell you which offer is real.

Who it's for

Teams about to sign an AI or software vendor on a deck and a demo, with no technical way to compare offers or catch lock-in.

What you get

Translate business goals into a precise, testable requirements spec (the RFP)
Score incoming offers against the spec, not the sales pitch
Surface hidden lock-in, integration risk and inflated estimates before you sign

Result promise

A defensible vendor decision in 3–4 weeks, with every risk in writing.

02 · Engineering DNA

How I work, in three habits.

I see the red the moment I walk in.

Within a week you stop hearing “we're on track.” You start seeing velocity, capacity and burn-rate, and the gap between them and the roadmap. I don't translate bad news into corporate language. If a project is red, you know it's red, with the specific reason and the cost of leaving it that way.

Your engineers ship more as the noise stops at me.

I absorb the stakeholder churn, the shifting priorities and the politics, and pass the team one clear set of expectations. They build; I handle the rest. The change you feel is quiet: fewer status meetings, fewer surprises, and work that reaches production instead of dying in review.

“Done” means it's live, monitored and documented.

I bring an Industrial Definition of Done from regulated, high-stakes environments. A model isn't finished when the demo impresses. It's finished when it runs in production, survives an EU AI Act review, and someone other than its author can maintain it. That bar is uncomfortable at first. Then it becomes the reason things actually ship.

03 · Proof

Shipped, not slideware.

MACHINE VISION / INDUSTRY 4.0

Challenge

A new MLOps platform had to reach production fast, with a web application teams could actually use to manage and deploy computer vision models.

Result

Production MLOps platform shipped in 6 months, with the web application live in users' hands.

“I owned the web application: the layer where an MLOps platform stops being infrastructure and becomes a product people actually work in.”

Banking / AML & compliance

Challenge

Manual KYC onboarding was slow, expensive and exposed to regulatory risk, with no production-grade platform.

Result

−80% onboarding time, −70% KYC operational cost, greenfield to production.

“Compliance and speed aren't a trade-off. They're both an engineering problem with the same answer: discipline.”

Fintech / fraud prevention

Challenge

Fraud scoring needed sub-second decisions at scale, with no room to slip the delivery date.

Result

<1s latency, delivered on time.

“Latency targets are a promise to the user. You design for them on day one or you never hit them.”