AI & Machine Learning Company in Stockholm

Three delivery shapes drive the range: a scoped AI proof of concept at Stockholm rates, a custom production ML model (fraud scoring, credit decisioning, recommendation, document extraction), or full production AI systems with RAG, multiple models, fine-tuning, and enterprise integrations. Scope drivers include data audit, baseline model, and hosted demo. Stockholm sits at the top of Nordic AI rates because of the Spotify, Klarna, and Truecaller talent premium, though typically below London on full-stack ML pricing. We give fixed-fee proposals rather than open T and M estimates.

The EU AI Act applies in stages: prohibitions on unacceptable risk systems live since 2 February 2025, general-purpose AI obligations from 2 August 2025, high-risk system obligations from 2 August 2026, and additional high-risk obligations through 2 August 2027. Our discovery phase classifies your system across the four risk tiers, runs an Article 35 GDPR DPIA where automated decision-making is in scope, and produces a Conformity Assessment plan for high-risk systems (Annex III categories including credit scoring, employment screening, education access, essential services, law enforcement, migration). High-risk builds ship with technical documentation under Article 11, a quality management system under Article 17, human oversight controls under Article 14, accuracy and robustness testing under Article 15, and post-market monitoring under Article 72. IMY enforces in Sweden alongside Finansinspektionen where finance touches AI.

Yes. We ship RAG systems, internal copilots, document extraction pipelines, and agent workflows for Nordic banks and fintechs. For SEB, Swedbank, Handelsbanken, Nordea Sweden, Klarna, and Avanza we stay inside EU regions (Bedrock and Azure OpenAI in eu-north-1 and eu-west-1, Mistral La Plateforme for EU-sovereign inference), apply Finansinspektionen outsourcing requirements and EBA Guidelines on ICT risk, and produce the model risk documentation internal audit needs. We have patterned deployments after public work from Klarna AI Assistant (the OpenAI-powered customer service rollout) and the AML transaction monitoring patterns Nordic banks share through the Nordic Financial CERT and EBA Q&A. Output logs feed existing SIEM and model risk platforms.

GDPR Article 22 restricts solely automated decisions producing legal or similarly significant effects, with carve-outs for contract necessity, explicit consent, or EU/Swedish law authorisation. Our discovery phase identifies any Article 22 trigger (credit decisioning under Konsumentkreditlagen, employment screening, insurance underwriting, essential service access) and designs in human review, the right to obtain human intervention, and meaningful information about the logic under Articles 13(2)(f) and 15(1)(h). For Klarna-style consumer credit we follow the Swedish Konsumentverket guidance, the EU Consumer Credit Directive 2023 obligations applying through 2026, and IMY's published views on profiling. Outputs include a Schrems II transfer impact assessment where US sub-processors are involved.

We default to AWS eu-north-1 (Stockholm region, lowest-carbon AWS region globally on Nordic hydro and wind power) for training and inference, with eu-west-1 (Dublin) as the cross-region replication and DR target. Azure Sweden Central (Gavle, opened 2021) supports Microsoft-aligned clients, and GCP europe-north1 (Hamina, Finland, seawater-cooled) covers Google standards. For LLMs that must stay in the EU we use Mistral La Plateforme, Bedrock with no-train Anthropic and Cohere terms, and self-hosted Llama 3.3 or Mistral on EU GPU instances. Cross-border to US regions is governed by an explicit Schrems II Transfer Impact Assessment under EDPB guidance, with the EU-US Data Privacy Framework as the standard transfer mechanism for certified US recipients.

A typical Stockholm fintech model (fraud scoring, AML transaction monitoring, credit decisioning, KYC document extraction) takes fourteen to twenty-four weeks from kickoff to production, assuming clean historical data and Finansinspektionen outsourcing documentation as part of scope. Week 1 to 4 is data audit, DPIA, and EU AI Act classification 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 FI conduct expectations. If you are pre-data or need labelling, add six to eight weeks. We have shipped to this cadence inside FI-supervised stacks alongside Klarna-style BNPL and SEB-style traditional banking architectures.

Yes, and it is one of the strongest reasons to host AI workloads in Stockholm. AWS eu-north-1 is one of the lowest-carbon AWS regions globally, running on Swedish grid power dominated by hydro and wind. Azure Sweden Central (Gavle) and GCP europe-north1 (Hamina, Finland) operate on similarly clean Nordic energy. For clients with CSRD sustainability reporting obligations or Science Based Targets initiative commitments, we deliver a Scope 3 carbon estimate per training run using the AWS Customer Carbon Footprint Tool, Microsoft Sustainability Manager, or Google Cloud Carbon Footprint, and design batch training to coincide with high-renewable hours on the Swedish grid. CSRD ESRS E1 climate disclosures get specific compute-emissions numbers, not a hand-waved estimate.

Yes, and we design for it on day one. Spotify, Klarna, Truecaller, and Storytel serve hundreds of millions of users across the EU, US, UK, Latin America, India, and Southeast Asia, which means EU AI Act, GDPR, US state laws (Colorado AI Act, California SB-1047 follow-ons, NYC Local Law 144), UK ICO, India DPDP Act 2023, and Brazil LGPD obligations stack. We architect models with regional data isolation, route EU customer inference through eu-north-1, route US through us-east-1 or us-west-2, and keep UK customers inside eu-west-2 (London) under the UK Adequacy Decision. EU AI Act risk classification ships in the model card from the start so a future Conformity Assessment is not a rewrite.