AI & Machine Learning Company in Amsterdam

Under the EU AI Act the risk tier of your use case does more to an Amsterdam ML budget than the model architecture ever will. The three options are a scoped AI proof of concept at Amsterdam rates over six to ten weeks, covering data audit, a baseline model, and a hosted demo, a custom production ML model (fraud scoring, churn prediction, document extraction, recommendation ranking) including MLOps, monitoring, and an EU AI Act-aligned model card, or a full production AI system with RAG, multiple models, fine-tuning, and enterprise integrations. Amsterdam rates sit above Rotterdam and Eindhoven because of the Booking.com, Adyen, and ASML talent premium. We give fixed-fee proposals rather than open T and M estimates, billed in EUR with Dutch BTW (21 percent) applied where required.

The EU AI Act entered into force in August 2024 with a phased timeline: prohibited practices applied from February 2025, general-purpose AI obligations from August 2025, and high-risk system obligations from August 2026 (with some product-safety integrated rules running to 2027). It classifies systems as prohibited, high-risk (employment, credit scoring, biometrics, critical infrastructure, education, law enforcement, migration, justice), limited-risk (chatbots, deepfakes), or minimal. Our discovery phase runs an EU AI Act classification on your use case, and high-risk builds ship with the Annex IV technical documentation, a conformity assessment plan, post-market monitoring, and human oversight controls. The AP is the lead Dutch market surveillance authority for many categories, and they have been signalling enforcement intent since the cookie wall ruling in 2019.

Yes. We ship RAG systems, internal copilots, document extraction pipelines, and agent workflows for Dutch banks and fintechs including patterns suitable for ING, ABN AMRO, Rabobank, Adyen, Mollie, and bunq stacks. For DNB-regulated clients we stay inside EU regions (Mistral Paris endpoints, AWS eu-west-1 or eu-central-1 Bedrock, Azure West Europe OpenAI), apply DNB model risk and DORA operational resilience controls, and produce the challenger model documentation internal audit needs. We pattern deployments after public work from Adyen on fraud and from Booking.com on recommendation, including strict PII redaction under GDPR, prompt injection defences, human-in-the-loop review gates, and evaluation harnesses that run before every production push. Output logs feed existing SIEM and model risk platforms.

Every Amsterdam engagement involving personal data ships with a Data Protection Impact Assessment aligned with Article 35 GDPR and the AP’s published DPIA list (which includes large-scale profiling, biometric processing, and systematic monitoring of public spaces). For high-risk processing we run a balancing test, document the legal basis, and design data minimisation into the model pipeline (feature ablation, k-anonymity where appropriate, differential privacy when scale warrants it). The AP is one of the strictest DPAs in the EU; they banned cookie walls in 2019 and have fined under GDPR aggressively. We default to assuming prior consultation with the AP may be required for high-risk pipelines and scope timeline accordingly.

When a project requires genuine research (novel model architectures, rare-event detection, reinforcement learning in production, causal inference at scale), we scope collaborations with Centrum Wiskunde & Informatica (CWI), the University of Amsterdam Informatics Institute, VU Amsterdam, or TU Delft labs rather than overselling in-house capability. CWI is one of Europe’s historic ML and theoretical CS centres, where Guido van Rossum built Python and where active ML research feeds Dutch industry. Our core team handles applied engineering, MLOps, and productionisation, which is where most Amsterdam AI projects actually stall. For standard work (RAG, fine-tuning, classical ML, computer vision on known architectures) no academic partner is needed. We will tell you up front which bucket your problem fits into.

We default to AWS eu-west-1 (Ireland) and eu-central-1 (Frankfurt), Azure West Europe (Amsterdam data centres, which we pitch for literal Amsterdam-local inference), and GCP europe-west4 (Eemshaven, NL, green-energy region) for storage, training, and inference. For LLMs that must stay strictly in the EU (DNB-regulated banking, AP-sensitive special category processing, public sector workloads) we use Mistral’s Paris endpoints or self-host Llama 3 and Mistral on EU GPU instances. Schrems II analysis informs every cross-border decision; transfers to US providers run under EU-US Data Privacy Framework adequacy with documented Transfer Impact Assessments. We document residency in the model card and the data processing agreement so AP, DNB, and internal audit have a clear answer.

A typical Amsterdam fintech model (fraud scoring patterned on Adyen-style risk, credit decisioning, KYC document extraction, AML transaction monitoring) takes fourteen to twenty-two weeks from kickoff to production, assuming clean historical data and DNB plus EU AI Act documentation as part of scope. Week 1 to 4 is data audit, DPIA, and baseline. Week 5 to 12 is modelling, iteration, and challenger testing. Week 13 to 18 is MLOps, monitoring, shadow mode deployment, and internal audit review. Week 19 onward is gradual rollout under DORA-aligned operational resilience controls. If you are pre-data or need labelling, add six to eight weeks. We have shipped to this cadence inside EU regulated stacks.

Most custom AI work qualifies for the WBSO (Wet Bevordering Speur- en Ontwikkelingswerk), the Dutch R&D tax credit administered by RVO (Rijksdienst voor Ondernemend Nederland). WBSO returns a payroll tax reduction of 32 percent on the first EUR 350,000 of eligible R&D wage costs and 16 percent above that (rates verify annually). Eligible activity generally includes novel modelling, architecture experimentation, algorithmic uncertainty, and systematic investigation, not routine integration. We deliver time-tracked engineer logs, technical narratives, experiment histories, and failed-hypothesis documentation aligned with the RVO application format. Amsterdam scale-ups can sometimes stack the Innovation Box (effective 9 percent corporate tax on qualifying IP income) on downstream commercialisation. Final eligibility sits with your WBSO consultant and RVO.