AI Models & Attribution
AINA publicly attributes all AI capabilities to the open-source models developed by the Barcelona Supercomputing Center (BSC) as part of Projecte Aina.
Barcelona Supercomputing Center (BSC) — Centro Nacional de Supercomputación
All primary AI models used by SEBA are developed and maintained by the BSC as part of Projecte Aina, a public initiative to build open-source AI for the Catalan language.
Salamandra 2 is a family of open-source large language models developed by the Barcelona Supercomputing Center (BSC) as part of Projecte Aina. Trained on a multilingual corpus with strong coverage of Catalan, Spanish, and English, it is the primary language model powering SEBA's conversational assistant (Aina), lesson planning, and curriculum-aligned content generation.
Languages
Licence
Apache 2.0Used in SEBA for
- Aina conversational assistant
- Lesson plan generation
- Worksheet and materials creation
- Curriculum-aligned question generation
- Bias detection and content moderation
Àguila (Agile General-purpose Language model for Iberian LAnguages) is an open-source bilingual language model developed by the Barcelona Supercomputing Center (BSC). It is optimised for Spanish and Catalan and serves as a complementary model in SEBA for tasks requiring high-quality Iberian language understanding, including reading comprehension, summarisation, and student feedback generation.
Languages
Licence
Apache 2.0Used in SEBA for
- Student feedback generation
- Reading comprehension tasks
- Summarisation of educational content
- Spanish and Catalan language understanding
Whisper is an open-source automatic speech recognition (ASR) model developed by OpenAI. SEBA uses it via the Hugging Face Inference API for voice transcription features, enabling teachers to dictate lesson notes and students to submit spoken answers.
Languages
Licence
MITUsed in SEBA for
- Voice transcription for lesson notes
- Spoken student answer processing
Compliance & Transparency
All AI models used by AINA are open-source, EEA-hosted, and subject to the platform Data Processing Agreement. Model outputs are subject to human-in-the-loop review by teachers before being used in student-facing contexts.