CIIIC

CIIIC is the Creative Industries Immersive Impact Coalition, a Dutch Nationaal Groeifonds programme running from 2024 to 2030 with a budget of €62.5 million. Its job is to strengthen the Dutch immersive experience sector: VR, AR, mixed reality and spatial computing, across research, talent, facilities and production. It is not a single institute but an umbrella coalition, coordinating funding and knowledge across universities, hogescholen, studios, labs, festivals, museums and individual makers.

This is my full-time job. I am responsible for the digital infrastructure and data side of the programme: the public website, the applications the programme team runs on day to day, and increasingly the AI tooling that makes a sprawling coalition legible to the people working inside it. I came into this from the Nieuwe Instituut, where I had been running online platforms since 2021.

Rebuilding the website on Payload and Astro

In 2026 I rebuilt ciiic.nl from scratch. The old site was a Craft CMS installation that had accumulated years of structural compromises; the new one is a monorepo with Payload CMS v3 as the content layer and Astro with server-side rendering as the front end, on PostgreSQL, with a SCSS design system and full NL/EN bilingual content.

The interesting part was not the stack, it was the editorial model. CIIIC runs ten funding schemes across five programme lines, each with its own deadlines, partners and audience segmentation. Most of the difficulty in a site like this is deciding what belongs in the CMS versus what belongs in code, and resisting the urge to build features the underlying data discipline does not yet justify. For the launch I parked an AI-powered wayfinding feature because the questions our audiences actually asked turned out to be finer-grained than anything the output could usefully answer. It is hidden, not deleted; it can come back once the events and news tagging is properly on its feet.

Twelve MCP servers

The programme generates a lot of state across a lot of systems: Notion for CRM, regelingen and milestones, Payload for published content, ImageKit for the image library, Gravity Forms for intake, Circle for community. ciiic-mcp is a monorepo of twelve Model Context Protocol servers, one per system, each independently runnable and each usable from Claude Code, Claude Desktop or our internal chatbot.

They follow a cascading retrieval pattern. A curated Markdown knowledge base comes first, because it is fast, structured and inter-linked; the other servers are for live data and write actions. Knowledge base pages point at the specific Notion databases behind them, so an agent knows where to look when it needs something more current than the summary.

The point of all this is not novelty. It is that colleagues who do not write code can ask a question in plain Dutch and get an answer grounded in what the programme actually holds, rather than a plausible-sounding guess.

The applications around it

A programme like this runs on a lot of small systems that nobody sees:

• dashboard.ciiic.nl is a SvelteKit application tracking the marketing and event pipeline: registrations, campaign metrics, mailings, survey results, pulled together from GA4, Circle, Gravity Forms and our own database.
• beeldbank.ciiic.nl is the image library and OG-image generator. It browses and filters the ImageKit collection, and generates social cards on demand through an API, so every announcement gets a properly branded image without anyone opening a design tool.
• monitor.ciiic.nl cross-checks funding scheme status across CMSes. forms.ciiic.nl, bot.ciiic.nl, docs.ciiic.nl and go.ciiic.nl handle intake, relaying, documents and short links.
• ai.ciiic.nl is a retrieval-augmented chatbot over the internal knowledge base, aimed squarely at non-technical colleagues.

Alongside the building there is a steady stream of migration work: Hubspot to a Notion-based CRM, Craft to Payload, Mailchimp segment restructuring, Microsoft Graph integration. Migration engineering has quietly become one of the skills I care most about getting good at, because it is where the gap between a clean architecture and a working organisation actually shows up.

What I am trying to get right

Two things, mostly. The first is knowledge systems that non-technical people can genuinely use, which is a documentation problem at least as much as a retrieval problem. The second is AI tools that do work rather than generate work about work: agents that close a loop and take an action, not ones that hand you a list of suggestions to process later.