Chiel Kas

I joined KVK to help launch their new real-time data-delivery platform for the Handelsregister (the Dutch business register), then stayed on to modernise how the engineering teams ship software.

Data-delivery platform:

• I built a custom Kafka-based replication mechanism that projects the Handelsregister into a fast, read-only store holding only current records — the source database is large and history-heavy, so direct queries are slow.
• I helped design the Kafka event messages and the reconciliation system that keeps the read model real-time and aims for strong consistency.
• I cut the initial full load of all KVK records from a projected several months to under a week.

Platform & delivery modernisation:

• I introduced GitOps and ArgoCD across KVK: ran a workshop for 100+ developers, set it up with the platform teams, and reworked the shared CI/CD pipeline templates.
• I migrated around twenty applications to ArgoCD continuous delivery, cutting their deployment from days of manual steps to ~30 minutes, fully automated.

AI-assisted engineering:

• I was closely involved in the internal community that used GitHub Copilot fanatically, and went deep on agentic coding myself. In June 2026 I gave an internal talk on the secure use of agentic-coding tools — it's where I think the work is going, and I want us ready for it.

Getting other teams moving — running the workshops, doing the convincing — is the work I keep coming back to.

I was one of four architects on this Bol.com spin-off, owning the software-engineering platform. I pushed hard for domain-driven design, event-driven architecture and zero-friction CI/CD, so a small team could keep shipping fast without the platform turning into a mess. GitHub Copilot was already part of my toolkit here; being early with that kind of tooling is a pattern of mine.

• In three months we went from nothing to live in production: real route planning, drivers with apps, actual parcels moving. I still think that pace is a little ridiculous.
• I set up the platform on GKE with GitOps through ArgoCD, and Terraform for infrastructure-as-code, so the whole environment was reproducible from scratch.
• I designed a zero-trust security model with Workload Identity, so services authenticated by identity rather than network location from day one.
• I helped design the real-time driver tracking: a GraphQL ingestion API feeding Google Pub/Sub, which the teams then built. Every truck's position showed live on the planning dashboard — watching them move across it never got old.
• We launched in summer, kept scaling, and the platform took its first Black Friday peak without a hitch.

What I'm proudest of is how fast we built something genuinely new — and that the quality held up, which the business noticed and valued. I also loved being close enough to talk to the drivers using the app and hear first-hand what happens out in the wild; moving things from A to B sounds mundane, but it's strangely satisfying.

I worked at BRMG — Bol Retail Media Group, the advertising side of bol.com — where I got a front-row seat on how the internet actually makes money. High tech, high volume, genuinely great people. Two projects stand out.

Sponsored-products platform:

• We built the platform sellers use to promote their products across bol.com.
• I built a new REST API for external advertisers, with Kafka event-streaming for real-time updates.
• I tuned the Kubernetes autoscaling so it held up through the busiest campaign weeks, and stood up the Prometheus/Grafana monitoring behind it.

Cookie Consent platform:

• I rebuilt the legacy system into a cloud-native service on Kubernetes — the official GDPR consent registry for every bol.com visitor.
• I brought response times down from 500ms to 50ms.
• I migrated 8 million daily users with zero downtime.

I became known as "Mr Cookie" while I was there. Nobody loves cookie consent, but somebody had to own it, and I didn't mind. The performance and zero-downtime numbers are the ones I'm proudest of.

I led R&D at this mobile-payments start-up: four remote teams across Serbia and Romania — two mobile, two back-end — plus the back-end API architecture and the operations team.

• I took deployments from weekly to several times a day with feature flags, automated testing and deploy automation. Getting a system to deploy in minutes instead of days is sort of my thing.
• I owned the back-end API architecture and put real work into the platform's performance — profiling and tuning until it held up.
• I built a serverless proof-of-concept on AWS Lambda that made it into production.
• I explored PSD2 for payment initiation, and we built on a major Dutch gift-card operator's wallet system.

Coaching the teams on engineering practices and watching their productivity climb is what made the role click for me.

EDIA is an education-technology software consultancy; I came in as a team lead and grew into lead engineer for the company — four development teams plus operations, around 20 engineers. I set the technical architecture and looked after knowledge sharing, developer productivity and software quality across the teams.

• This was the deep-learning wave. I worked at the architecture level alongside the team's TensorFlow/PyTorch work, and moved us onto AWS, Kubernetes and microservices.
• Beyond education, I built an e-healthcare integration on Koppeltaal and FHIR.
• I led two teams on a proof of concept for an AI-supported learner-management system in Abu Dhabi.

The Abu Dhabi project was a bit mad, honestly — but being in it with the teams is something I'm proud of.

Energyworx builds a meter-data-management platform for the energy and utilities sector. This was my move from consulting into an in-house product team, and my first proper cloud work — on Google Cloud (App Engine), with Cassandra.

• I migrated the platform's Cassandra store from on-prem into Google Cloud.
• I helped the team mature its continuous delivery.
• In-house I focused on the team's development productivity; externally I helped clients integrate the Energyworx REST API into their own systems.

A year on ABN AMRO's internet-banking front-end: first leading a front-end team, then advising the bank on faster front-end development.

Team lead:

• I led a 6-person scrum team rebuilding the rights- and identity-management front-end: pure HTML/CSS/JS single-page apps on a REST backend.
• I owned the front-end build, and migrated it from the legacy Maven setup to Grunt/Gulp.

Prototype & workflow:

• I built an architectural prototype for the internet-banking app on Backbase and AngularJS.
• I advised on the front-end developers' workflow; we brought the release cycle from three months down to one.

These were wild days for JavaScript — a new framework seemingly every week, peak Angular hype. I came in as a backend developer and held my own in a front-end field that was maturing fast.

Two Capgemini engagements on the Java-based Nuxeo ECM platform.

NFI (Netherlands Forensic Institute):

• I was technical lead of the team that built a digital case-management system for forensic investigations, used in real casework. It was Capgemini's first Nuxeo ECM implementation in the Netherlands.
• I set up the team's CI/CD with Jira, Jenkins and Git, and coached the other NFI IT teams to adopt continuous deployment.

MVB (Frankfurt — distributor of all German books to retailers):

• I advised their team of experienced Java developers, ran a small proof of concept, and recommended Nuxeo ECM for their 25 TB catalogue of e-books and audiobooks.

This is where my DevOps streak really started — an early Docker adopter, running Nuxeo locally for dev and test back in 2013.

Two short Capgemini assignments, both around enterprise content and search.

• At UWV, I built a migration tool that moved a legacy "fast" digital archive of over 100 million documents into IBM FileNet, and advised the team on continuous integration.
• At BZK (Ministry of the Interior), I tested an internal news-aggregation and search system that indexed news sources and alerted people to new articles in their areas of interest.

My first two Capgemini assignments, both implementing Oracle UCM for enterprise content and document management.

KPN:

• I integrated Oracle UCM into KPN's invoicing process, generating and archiving the invoices as PDF.
• I optimised invoice generation from 5k to 50k invoices per hour.

ICTRO (IT services for the Ministry of Justice):

• I migrated the document-management layer of GPS — the judiciary's case-handling system — from Hummingbird to Oracle UCM.

After my bachelor project I stayed on at ING, building content-management web applications for the dealing-room back office — a business-critical Java/EMC Documentum environment. I worked with end users and a business analyst to turn requirements into designs, and started assigning and reviewing other developers' work.

• I automated a contract-chasing process with Documentum workflow: it chased counterparties by fax or email on its own. Faxes were generated on the fly and sent via a third-party web service.
• I built a Java Struts application for the security operations department to streamline Documentum user management. It tied into the company-wide LDAP directory and could export users and their roles to Excel.

My first job in tech: a traineeship at ING combined with my bachelor study — four days a week at the bank, one day studying at the Haagse Hogeschool. I started in support and moved into building software, which was where I wanted to be.

• I provided third-line support for around ten of ING's custom EMC/Documentum applications.
• I took on Java/Documentum development as projects came up.
• For my bachelor project I led the technical side of a desktop-to-web migration: a Struts web front-end over a fast-growing archive of more than 5 million contracts. I used Ajax to keep the responsiveness of the old desktop client — back when nobody called it a single-page app yet.
• I built a low-volume scanning application on Input Accel with a Visual Basic front-end.

A bachelor project usually ends up on a shelf; this one shipped and stayed in use. That's where I caught the bug for building software, and I never really stopped.