Engineering AML & KYC Compliance at Scale for a Leading Multinational iGaming Company

Project at a Glance

• Sector: iGaming

• Engagement: 2020–2021

• My Roles: Senior Technical Consultant, Engineering Manager, Project Manager & Data Architect

• Scope: Multi-region, multi-cloud Kubernetes platform running Kubeflow for ML/DL-driven AML and KYC compliance

• Infrastructure: AWS, GCP and on-premises, spread across multiple global locations

• Disciplines: Project Management, Lean Management, Mission Command, DevOps, MLOps, SecOps, Data Architecture, Data Engineering, Container Orchestration, Big Data & Data Analytics

Introduction

Between 2020 and 2021, I worked on one of the most demanding compliance-engineering programs of my career: building the data and machine learning infrastructure behind a new Anti-Money Laundering (AML) and Know Your Customer (KYC) capability for a leading, multinational iGaming company.

The Client. A multinational gaming-technology and iGaming operator with a regulated presence across numerous European markets the kind of organization most people in the sector would recognize immediately. I’m keeping the name out of this write-up out of respect for the confidentiality of the engagement.

The work originated inside the client’s SecOps organization, specifically a SMART AML and CFT Support Team (Combating the Financing of Terrorism), and it sat at the intersection of three disciplines that rarely share a roadmap comfortably: financial-crime compliance, enterprise data engineering, and large-scale Kubernetes infrastructure. I worked across all three first inside the Project Management Office (PMO) shaping the compliance framework, and later in a combined engineering, project management, and data architecture role leading the team that built it.

The Challenge

iGaming is one of the most heavily regulated digital industries in Europe, and by 2020 amid a wave of tightening EU anti-money-laundering directives the bar for AML and KYC compliance was rising fast for every operator in the sector. The specific problems we had to solve:

• Regulators increasingly expected real-time, auditable, explainable compliance decisions, not periodic manual reviews bolted onto legacy systems.

• The client operated across multiple regulated markets, so the solution had to hold up to scrutiny from several regulators at once, not just one home jurisdiction.

• Existing onboarding and transaction-monitoring processes were fragmented and largely manual, with no realistic path to scale alongside the business.

• Meeting the requirement meant standing up production-grade Machine Learning (ML) and Deep Learning (DL) workloads an MLOps problem as much as an infrastructure one, needing reproducible pipelines and model governance that both data scientists and AML analysts could trust.

• The underlying IT infrastructure had to be highly available, secure, and resilient to any single vendor’s outage spanning AWS, GCP, and on-premises data centers across two EU locations, multi-cloud and multi-region from day one, built as infrastructure-as-code.

• All of it had to be delivered by a fully remote, asynchronous, multinational team spanning engineering, legal, AML, and compliance specialists, while the client’s own legal and AML leadership were going through organizational change which kept requirements moving under our feet.

The Solution

I worked this program in two phases.

Phase One: Shaping the Compliance Framework

As a Senior Technical Consultant inside the PMO, I worked directly with business, AML, legal, and compliance specialists to translate regulatory intent what KYC and AML actually had to mean operationally for this business into a project framework engineering could build against. This is the unglamorous but make-or-break part of any compliance and ML program: getting compliance and engineering speaking the same language before a single pipeline gets built.

Because regulatory expectations and business priorities continued to evolve throughout the engagement, we deliberately applied Lean delivery principles rather than attempting to freeze requirements up front. Working in short feedback cycles with compliance, legal, and engineering stakeholders allowed the project to continuously refine priorities while maintaining a stable architectural direction, reducing rework and ensuring the platform remained aligned with changing regulatory and business needs.

Read more: The Art of Defining Project Goals

Phase Two: Building the Kubernetes and Kubeflow Platform

As the program moved into implementation, I transitioned into a combined Project Manager, Engineering Manager, and Data Architect role, leading a team of MLOps and DevOps engineers responsible for the platform’s technical delivery.

Because the engineering organization was fully remote and distributed across multiple countries and time zones, day-to-day execution relied heavily on a Mission Command style of leadership: clearly communicating intent, desired outcomes, architectural guardrails, and compliance constraints while empowering engineers to make decentralized technical decisions within well-defined boundaries. This approach increased delivery speed, reduced coordination overhead, and enabled teams to respond quickly to changing technical and regulatory requirements while remaining aligned with the program’s objectives.

The core of the solution was an end-to-end Machine Learning workflow built on Kubeflow, running on a highly available Kubernetes platform spanning two European locations to provide resilience and high availability. This was a production-grade, multi-cloud MLOps platform rather than a proof of concept, designed to support the client’s long-term AML and KYC compliance strategy.

Read more about Kubernetes: Introduction to Kubernetes

Data Architecture, Engineering, and Analytics

My primary technical focus was Data Architecture and Data Engineering: designing the data pipelines and analytics layer that transformed KYC onboarding information and transaction histories high-volume, multi-source, Big Data-scale datasets into reliable inputs for machine learning and deep learning models supporting identity verification, risk scoring, and transaction-pattern analysis, while ensuring every stage remained fully auditable for regulators.

Working within the client’s SecOps organization meant security was treated as a core architectural principle rather than an afterthought. Access control, encryption, audit logging, and infrastructure-as-code practices were incorporated from the outset. At the same time, the platform was engineered to operate consistently across AWS, GCP, and on-premises infrastructure, making multi-cloud portability, standardized deployment processes, and operational consistency just as important as the ML pipelines themselves.

The Result

By the time my engagement wrapped, the platform had moved from framework and architecture design into a working, production-track Kubernetes and Kubeflow environment running real ML and DL workloads for the AML and CFT team a genuine end-to-end pipeline from data ingestion through model inference.

What the program delivered:

• A reusable, multi-region, multi-cloud Kubernetes foundation the AML/CFT team could keep building on, rather than a one-off integration.

• A repeatable MLOps process, via Kubeflow, that let data scientists and AML analysts iterate on detection models without re-engineering infrastructure every time.

• A documented compliance framework connecting legal and AML requirements to concrete technical and data requirements the piece that made the engineering work defensible to regulators and auditors.

At the time, this was one of the largest AML/KYC machine-learning infrastructure programs the European iGaming sector had undertaken a real-world proof that Kubernetes- and Kubeflow-based MLOps could move from experimental to production-grade inside a heavily regulated, multi-jurisdictional environment.

Key Takeaways and Executive Insights

• Compliance is now a data and ML-engineering problem, not just a legal one. Bolting AML/KYC checks onto legacy systems doesn’t scale; regulators increasingly expect auditable, explainable, model-driven pipelines.

• The hardest part usually isn’t the Kubernetes cluster it’s the translation layer. Turning shifting legal and compliance intent into stable engineering requirements, fast enough to keep pace with organizational change, is where these programs succeed or fail.

• MLOps discipline is what makes an AML platform durable. Without a framework like Kubeflow, every model update becomes its own infrastructure project instead of a routine iteration.

• Multi-cloud, multi-region architecture is a requirement, not a luxury, for regulated multinational platforms for resilience, and for satisfying regulators across jurisdictions simultaneously.

• This is exactly where fractional, senior technical leadership earns its keep. Programs like this live or die on the connective tissue between business, legal, and engineering bridging that gap fast, without adding permanent headcount, is the core value a fractional CTO or Data Architect brings to a build like this.