From multilingual data to sovereign AI infrastructure

Multilingual foundations and statistical machine translation

Pangeanic began by managing multilingual content and structured bilingual assets at industrial scale. That experience led naturally into parallel corpus creation, translation memory reuse, terminology management, and statistical machine translation.

Work with large bilingual datasets and early open-source MT technology established data engineering practices that later became essential for neural models and modern AI training.

Industrial scaling, neural systems and European infrastructure

As machine translation moved from statistical architectures toward neural systems, Pangeanic expanded from individual engines to larger infrastructure, model customization and secure institutional workflows.

The European iADAATPA / MT Hub project created a secure, provider-neutral routing layer that allowed public administrations to access and compare several translation engines through a common infrastructure.

Neural scale, multilingual datasets, privacy and evaluation

Pangeanic coordinated NTEU — Neural Translation for the European Union, which delivered 552 direct neural translation directions covering every directed combination among the 24 official EU languages without requiring English as an intermediate pivot.

The company also coordinated MAPA, developing multilingual anonymization resources and technology for sensitive administrative, legal and medical content.

This period also consolidated work in neural adaptation, automated quality evaluation and the technologies that underpin Deep Adaptive AI Translation and Machine Translation Quality Estimation.

AI Data Operations, evaluation and sovereign AI

Pangeanic's data operations expanded beyond language pairs into training, evaluation and alignment data for modern AI systems.

Our work now spans multilingual text, speech, audio, images, video, multimodal datasets, expert feedback and model evaluation.

Collaboration with the Barcelona Supercomputing Center connects this trajectory with multilingual model training, evaluation, human feedback and European sovereign language models.

The current focus brings together Data for AI, AI Data Operations, evaluation and alignment, and sovereign AI system design.

Datasets for AI

Ready-to-license and bespoke assets across text, speech, audio, image, video, document intelligence and domain-specific evaluation.

AI Data Operations

Managed workflows connecting governed data, human review, feedback, evaluation, alignment, privacy and continuous quality improvement.

Evaluation & AI Quality Assurance

Multilingual benchmark design, model testing, scoring, failure analysis, regression testing and expert validation.

Model Alignment & RLHF

Human feedback, preference ranking, specialist review and multilingual supervision for controlled model behavior.

Task-specific and sovereign AI

Specialized models, knowledge grounding, multilingual assistants and controlled deployment across private infrastructure.

Ana Belén Fernández Bosch

Coordinates operations, human resources and AI data delivery, connecting multilingual contributors, quality processes and enterprise data programs.

José Miguel Herrera Maldonado, PhD

Leads machine learning work spanning speech resources, information retrieval, evaluation, multilingual data spaces and production-oriented AI systems.

Amando Estela

Provides technical continuity across infrastructure, projects and operations, with a long history in Pangeanic's machine translation and European technology programs.

Why Multilingual AI Data Quality Is Hard to Get Right

Why multilingual datasets require language-aware sourcing, provenance, native expertise and separate evaluation thresholds.

From Fine-Tuning to Red Teaming

The data, feedback, adversarial testing and regression cycles behind reliable production AI.

From Small Models to Sovereign AI

How data, ontologies, specialized models and operational control combine into practical sovereign AI architectures.