MEVO is a project to train a domain-specific large language model from scratch for digital marketing advisory, campaign analytics, SEO, paid media optimisation, web development support, and multilingual business communication in Nordic markets. The project is motivated by preliminary internal evidence from approximately 700 real-world Nordic digital marketing customer cases, which indicates that a specialised model can outperform generic large language models on customer-specific advisory tasks that require domain context, regional language variation, actionable prioritisation, and business-aware reasoning.
The proposed work will develop a fully owned foundation model and post-training stack under European control, including dataset curation, tokenizer design, pretraining, domain adaptation, instruction tuning, tool use, and evaluation. A core objective is to build a model that is more useful, more controllable, and more auditable than general-purpose open-weight models for professional advisory use. In particular, the project addresses practical limitations of current open-source models, including opaque data provenance, uneven Nordic-language performance, and difficulty in systematically auditing and mitigating demographic and domain-specific bias.
The project will follow a staged compute programme from small-scale screening models to large-scale pretraining, with the final outcome being a production-oriented MEVO foundation model optimised for Nordic digital marketing workflows. The resulting technology will strengthen European AI sovereignty, reduce dependency on third-party frontier models, and create a high-value, domain-specialised AI asset with full internal ownership of the model, training pipeline, and resulting intellectual property.
Principal Investigator, Company and Country
David Verterlund, Nordic Amazing Group AB, Sweden