Regional Language Research
Language systems frequently overlook the structural realities of Southeast Europe. Serbian, Croatian, and Bosnian share extensive linguistic foundations while retaining distinct standards, vocabularies, orthographies, and cultural context. Mijatovic focuses directly on these languages rather than treating them as generic translation targets.
Trained from Scratch
Adriaticum was trained from random initialization rather than fine-tuned from an existing base model. Research focuses on independent model lineage, direct training data curation, and architectural transparency.
Transparent Provenance & Engineering
Responsible engineering requires clear boundaries and documented lineage. Pretraining corpora, regional adaptation sources, and instruction data are methodically cataloged, with licensing notices published openly. We favor verifiable, documented systems over ungrounded claims.
Project Surfaces
Mijatovic develops cohesive surfaces around its core language research, each with deliberate boundaries and purpose.
Adriaticum
The core language model developed for the regional linguistic context of the Balkans.
Conversational Workspace
The primary interface for interacting with Adriaticum, designed for contextual dialogue and visible sources.
Platform & Developers
Developer console in beta for credentials and usage. Public API execution is not currently available.
Security & Provenance
Documented security baseline, owner-scoped data isolation, and comprehensive third-party notices.