AI & Machine Learning Company in Zurich

The core-banking systems a model must reach set the Zurich number: a scoped AI proof of concept at Zurich rates, a production ML model (KYC document intelligence, claims triage, churn, demand forecasting, hearing-loss screening), or full production AI platforms with RAG, fine-tuning, multiple models, and Avaloq, Temenos, or Guidewire enterprise integrations. Scope drivers include data audit, a baseline model, and a hosted demo on eu-central-2 or Switzerland North.

The revised Federal Act on Data Protection (nFADP / nDSG) entered into force on 1 September 2023 and is broadly GDPR-equivalent: Switzerland holds an EU adequacy decision (last reviewed 2024), data subjects have access, rectification, and portability rights, controllers must keep a processing record, and a DPIA is required for high-risk processing including most automated decision-making. Article 21 nFADP mirrors GDPR Article 22 on automated individual decisions: data subjects can demand human review. Differences worth flagging: the nFADP has no GDPR-style four-percent global-turnover fine, instead applying fines up to CHF 250,000 on individuals (not entities), and breach-notification timelines are softer ('as soon as possible' rather than 72 hours). Our DPIAs satisfy both regimes so EU-export clients do not run two documentation streams.

FINMA Circular 18/3 (Outsourcing - banks and insurers) governs how Swiss banks and insurers outsource significant business activities, which expressly includes cloud computing and AI services. Required controls include an inventory of outsourced services, a materiality classification, a written outsourcing agreement covering audit and FINMA access rights, a business-continuity plan, and ongoing risk management. AI-specific overlays come from FINMA Circular 23/1 operational risk and resilience, and FINMA's 2024 risk monitor explicitly on AI. Our regulated builds produce the FINMA Outsourcing-Notification documentation, the audit-rights clauses, and the exit strategy your internal audit and FINMA-supervised entity governance committee need. AWS eu-central-2 and Azure Switzerland North hold FINMA-accepted outsourcing patterns documented through their compliance offerings.

Yes. We ship document intelligence for client onboarding (KYC, KYB, suitability assessment, MiFID II suitability and Swiss FinSA equivalence), AML and sanctions transaction monitoring, portfolio-construction and rebalancing AI, ESG screening, and client-advisor copilots. Stack patterns integrate with Avaloq, Temenos T24, Olympic Banking System, and Murex. FINMA Circular 18/3 outsourcing classification and FINMA Circular 23/1 operational resilience are scoped at discovery. Models train and infer inside AWS eu-central-2, Azure Switzerland North, or Swiss-resident GPU clusters at Equinix and Interxion Zurich, with documented data flows that respect Swiss banking secrecy (Art. 47 Banking Act) and FADP Article 9 health-data carve-outs where life-insurance underwriting is in scope.

Regulation (EU) 2024/1689 applies extraterritorially under Article 2: an AI system put into service in the EU, used by a provider or deployer established in the EU, or whose output is used in the EU, falls inside scope regardless of where the provider sits. That captures most Swiss bank, insurer, pharma, and medtech AI because outputs flow to EU customers, EU subsidiaries, or EU regulators. Our discovery phase runs a dual FADP and EU AI Act classification, and high-risk Annex III builds (credit scoring, employment, biometric identification, critical infrastructure, education access) ship with Annex IV technical documentation, a quality management system, post-market monitoring, and a designated EU authorised representative when the provider is established outside the EU. Swiss-only deployments (e.g. cantonal public services with no EU output) get FADP-only treatment.

We default to AWS eu-central-2 (Zurich), Azure Switzerland North (Zurich) and Switzerland West (Geneva), and GCP europe-west6 (Zurich) for storage, training, and inference. For LLMs that must stay in Switzerland we use Azure OpenAI Switzerland North, Anthropic on Bedrock eu-central-2, or self-hosted Llama 3.3 and Mistral on H100 clusters at Equinix Zurich (ZH4, ZH5) and Interxion Zurich. Cross-border to EU regions (eu-central-1 Frankfurt) is acceptable for non-banking-secrecy workloads under the EU adequacy decision and a documented Auftragsverarbeitungsvertrag. Cross-border to US is acceptable only after a Transfer Impact Assessment, a Data Privacy Framework-certified recipient, and a FINMA outsourcing notification where applicable. Residency is documented in the model card, FADP processing record, and Auslagerungs-Inventar.

When a project requires genuine research (novel architectures, optimal transport, geometric deep learning, theorem-proving AI, neuroscience-grade brain-computer interface work), we scope collaborations with ETH AI Center, EPFL CIS, or the University of Zurich AI lab rather than overselling in-house capability. Our core team handles applied engineering, MLOps, and productionisation, which is where most Zurich AI projects actually stall, particularly when an ETH or EPFL spinout (Cevotec, MindMaze, Daedalean) hits the production-readiness wall. 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 and have introduced Zurich clients to ETH AI Center industry-engagement programmes when budget and roadmap fit.

Yes. We have scoped clinical AI for trial-recruitment, real-world evidence, dermatology, hearing-loss screening, neurorehabilitation, and surgical-robotics decision support. Where the software is positioned as a medical device, we run the build through Swissmedic Medizinprodukteverordnung (MepV) Class IIa or IIb, Heilmittelgesetz (HMG), and EU MDR Regulation (EU) 2017/745 under the Mutual Recognition Agreement framework (currently lapsed for medtech since 2021, so EU CE marking requires an EU-based Authorised Representative). IEC 62304 software lifecycle, ISO 13485 quality management, ISO 14971 risk management, and ISO 14155 clinical investigation feed the conformity assessment. FDA 510(k) is scoped separately where US distribution is in scope. All training and inference stay inside Switzerland or EU regions under FADP Article 6 (formerly Article 3) special-category handling.