Quilyx

Programme leadership
and AI strategy

We take AI ambitions from the boardroom to working results.

See the systems ↓

The claim

What we claim

Quilyx is a Dutch IT consultancy, based in Utrecht, The Netherlands.

Our claim is narrow and testable. We take AI ambitions from the boardroom to working results, and our own firm is the proof: we build systems that take over the support tasks, so senior people spend their hours deciding instead of processing. Augmentation, not automation.

AI does not replace programme judgment, it enhances it.

Sovereignty by design: sensitive client and supplier data is handled only by our own AI, on our own hardware in the Netherlands, with local model inference. No external AI service ever sees it. GDPR and the EU AI Act are design inputs from the start. For our own knowledge work we also use hosted models, and this site was built with them.

The proof

Five systems, each at its own stage

What runs the firm

Quilance

In production · internal use

Our placement-automation engine. It reads inbound assignment streams, splits and triages them, extracts each structured request, and matches it against registered freelance professionals.

01Intake Marketplaces, desks, mail 02Filter The stream narrows 03Triage Fit or no fit, with reasons 04Extract Structured fields 05Match Best fit 06Place The placement contract Freelyx Portal
01Intake. Assignment streams arrive from marketplaces, sourcing desks and mail.
02Filter. The inbound stream narrows: multi-role requests become single requests.
03Triage. Fit or no fit, fast, with reasons attached.
04Extract. Rate, location, deadline, skills: structured fields instead of prose.
05Match. Ranked against the registered professionals.
06Place. A human decides. The system prepared.
·Freelyx. Registered professionals flow in through our placement portal.
Runs on our own hardware Local model inference Never fed to external AI

Before its own launch it surfaced our first major result-shaped assignment.
Its public face is Freelyx, our placement portal.

QuiDAS

Working engine

A discovery engine that maps where demand surfaces: the procurement portals, dynamic purchasing systems and supplier desks behind which public and private buyers publish their needs.

QuiDAS finds the portals, removes the duplicates and aggregators, and turns the landscape into a worklist.

1,535raw references processed
39distinct systems still to register
176already connected
34live systems with no portal at all

As of 9 October 2026 · dots illustrative, counts real

A working engine in daily internal use.

Freelyx

Live

Our freelancer placement portal: the public, logged-in face of the Quilance engine. Registered professionals sign in to read the assignments Quilance has surfaced, split and structured.

The portal never holds the link between an account and a person. That link stays on our own hardware, so a portal breach cannot hand anyone a list of names.

01Export From Quilance 02Lock Sealed and anonymised 03Read Two-factor optional
01Quilance seals an anonymised assignment export on our own hardware.
02The portal receives it. Two-factor authentication can be switched on for every account.
03A signed-in professional reads the assignment. Nothing is readable without signing in.
No path back. The portal cannot reach our systems, so a portal breach stays a portal breach.
Pseudonymous accounts Two-factor available One-way feed

Signing up is free and open through a short form, and a person reviews every application before portal access.
Assignments never appear on a public page. Reading them requires signing in.

Prototypes and early versions

QEPMS

Prototype · synthetic data

A portfolio-intelligence layer for enterprise change. It reads the sources a portfolio already has, normalises them into one model that does not care which delivery method produced them, and emits a single overview with citable board packs, a map of what it does and does not cover, and flags where the sources contradict each other.

Portfolio reporting fails on the feeding, not the reporting. So this one reads the artefacts people already produce instead of asking them to maintain another tracker.

Prototype board · synthetic data · drawn from the working demo

The core is built and tested, entirely on synthetic portfolio data.
A demo is available, running on our own local hardware.

I.R.E.N.E.

Early version running

I Remember Everything, No Exceptions

A digital mirror of a physical brain. What the head files away in long-term memory, the mirror keeps top of mind: current, sourced and one question away, without the head having to carry it. Its design follows how a brain divides work: every compartment has its own store and its own agent, one orchestrator holds them together, and as it grows each compartment gains subagents under its own orchestrator, with one above them all.

The payoff is headroom. A head that does not have to keep everything top of mind is free for the work that needs it whole, for example for running programmes. That is how AI supports the programme manager here.

01Capture Raw thought, verbatim 03Projects In flight, one next step 05Commitments Dated, met or open 02Facts Verified, sourced, dated 04Decisions Chosen, and superseded 06Shielded Closed by design
01Capture. Raw thought in, verbatim, never tidied on the way.
02Facts. Verified, with a source and a date attached.
03Projects. Everything in flight, each carrying one next step.
04Decisions. What was chosen, and what it superseded.
05Commitments. Dated, and either met or still open.
06Shielded. Closed by design. What is in there is not shown here, or anywhere.
Provenance on every fact Evidence threshold Contradiction flags

A first version already runs inside the firm every day, on hosted models.
The version that runs on our own hardware is next.

The approach

How we run AI

Humans decide, systems prepare

Every output lands in a human workflow. Nothing auto-commits, nothing auto-sends.

Sovereignty by design

Client and supplier data is handled by our own AI on our own hardware in the Netherlands. European data law is a design input, not an afterthought.

Build small, run real

A system earns a bigger job by doing a small one in daily operation first.

This site works the same way: static files, no trackers, no third-party requests, built by AI agents and reviewed by the people below. The colophon has the receipts.

The founders

Who is behind it

Jeffry Turfboer and Bjorn van Rosengarten, founders of Quilyx

Jeffry Turfboer

Founder · Channel 01

Programme leader. More than twenty-five years of enterprise IT delivery. Claude Certified Architect (2026). Designs the systems on this page and builds them together with AI.

Bjorn van Rosengarten

Founder · Channel 02

Public-sector executive and programme leader. Former CTO at the City of Rotterdam. Led the Dutch municipalities’ information-security service (IBD). Managing director of Quilyx.

Utrecht · NLTwo foundersQuilyx B.V.

Member of NVIDIA Inception Registered Partner in the Claude Partner Network (CPN)
NVIDIA Inception Program member badge

Program memberships, not endorsements. The NVIDIA Inception badge is shown under the program’s member guidelines.

Contact

Talk to us

One address, no forms, no funnels.
It reaches the founders directly.

ai@quilyx.com