Build a representative Greek test set
Include domain terminology, names, accented text, scanned documents and realistic user questions.
Priority focus
AI that understands the Greek context.
Language quality, local knowledge and reliable evidence belong in the same evaluation. A fluent Greek answer still needs to be correct.
Test grounding and uncertainty
Check whether answers follow the supplied sources, remain current and acknowledge missing evidence.
Judge usefulness with real users
Compare task completion, reading clarity and correction effort across different user groups.