The model should help separate questions, interpret answers, explain marks, and teach the theory underneath. The specs and mark schemes still need to be stored as the app's source of truth.
Split the upload into parts, identify marks, and connect each part to the board, tier, and topic.
Store typed, drawn, or uploaded responses part-by-part so marking stays anchored to the right section.
Use stored specs and mark schemes first, with AI helping interpret methods and explain feedback.
Return marks, feedback, accepted methods, and a short revision note on the underlying theory.
Store official specs, topic maps, and mark-scheme guidance instead of expecting the model to remember everything.
Treat Edexcel as the active truth set in the first version, then add AQA and OCR separately.
Use AI to extract, classify, explain, and draft feedback, while keeping final marking logic grounded in board-aware data.
The best immediate path is: upload a question, separate the parts, capture the answer, then return marks and theory notes in one review session.