Educational AI evaluation
We evaluate explanations, source grounding and age-appropriate learning support through Pebble Bench.
ResearchLearning in low-connectivity environments
Serving educational AI where reliable cloud connectivity cannot be assumed. Pebble Local.
ResearchCurriculum-aligned knowledge and retrieval
Curated educational knowledge and curriculum resources that answers can be grounded in. Pebble Knowledge.
BuildingSpecialised educational models
Educational AI models and experiments, with published model cards. Pebble Models.
ResearchTools for educators
Technology supporting educators and learning environments. Pebble Classroom.
BuildingExplore the work behind accessible learning. Connect with our research priorities and contribute knowledge, code or classroom experience.
Pebble Bench — evaluation and benchmarking for educational AI
A benchmark for how educational AI explains, grounds and protects. Our focus: clear explanations, source grounding and learner safety.
Pebble Knowledge — curriculum-aligned knowledge corpus
Curated educational knowledge and curriculum resources, with licensing and provenance recorded per source.
Pebble Local — offline deployment reference
A reference deployment for a school-local Pebble node and the devices it serves.
Pebble Models — specialised educational models
Model cards cover purpose, supported domains, languages, evaluation, known limitations and safety.
Pebble Classroom — technology supporting educators
Built with educators, not around them.
A Pebble model card accommodates purpose, supported domains, languages, context, evaluation, known limitations, safety and data disclosure — and links to the report, repository and related research.
Safety is part of the product. Safeguarding, responsible-AI review and published model cards come before release, not after.
Education is something we build together.