Introducing Cereby Tutor — Your AI Study Partner Inside Every Quiz
Problem and motivation
Students often complete items from quizzes generated from their own notes or readings, then need to reconcile an incorrect response with the underlying material. The limiting factor is frequently locating relevant evidence in the original document, not a lack of effort. Generic chat assistants lack the quiz attempt context (current item, answer key, submission state) and typically do not have access to the same source file the quiz was built from.
Cereby Tutor addresses this by embedding a tutor panel in the quiz experience (Open quiz Cereby in the quiz modal). The system is aware of the active quiz, the current question index, and—where applicable—the resource associated with quiz generation. The design goal is to support formative practice: assistance during solving, then structured explanation after the official result is shown.Comparison with general-purpose assistants
A standalone chat model may discuss a topic broadly but does not automatically receive the structured quiz payload (stem, options, official explanation, and—after reveal—whether the learner was correct). Cereby Tutor is bound to the attempt: responses are scoped to the active question, use quiz metadata (title, topic, difficulty when available), and follow rules that preserve assessment integrity before submission.
Sessions are temporary and oriented to the quiz flow rather than long-lived cross-topic conversation elsewhere in the product.
Behavior prior to answer reveal
Until the quiz interface has revealed the official answer for the item, the tutor follows the Quiz help policy: it may provide hints, terminology clarification, and approach guidance without disclosing the correct response. This is intended to reduce answer leakage while still supporting learners who are stuck on reading or strategy.
Permitted support includes, for example:
- Clarification of terms in the stem
- High-level solution strategies (e.g., identifying negation, eliminating inconsistent options)
- Connection to optional learning path reading text when that excerpt is supplied as context
The learner remains responsible for the final selection; the model does not substitute for that step.
Behavior after answer reveal
Once the UI has shown the official answer (and any bundled explanation), constraints relax. The tutor may:
- Explain why the keyed response is correct.
- Contrast the learner’s submitted choice with the answer key when the attempt was incorrect, using metadata supplied for that question rather than inferring from chat alone.
- Provide concise or expanded explanations as requested, including mathematical notation or structured markdown when appropriate.
Replies are prefixed with the active Question N: identifier so the thread remains aligned with the on-screen item.
Source-grounded explanation
When a quiz is created from or linked to a resource in Cereby, Cereby Tutor can receive source context derived from that material. After reveal, learners may ask where the text supports the correct answer, how their response diverged from the source, or for quotations or paraphrases tied to specific passages.
Explanations can therefore be anchored in the same document used to generate or study for the quiz, rather than in unrelated web content. Source-grounded citation is conditional on that linkage; quizzes without an attached resource still receive explanation from the question stem and official explanation fields.
Implementation notes for users
- Per-question threads: Conversation history is maintained separately per question index. Clear resets the thread for the current item only.
- Learning path reading: When present, reading excerpts may be included in context to align explanations with assigned material.
- Formatting: Responses may use markdown, including math and diagrams where useful, consistent with other Cereby chat surfaces.
Summary
Cereby Tutor connects assessment, feedback, and—when applicable—primary sources in one workflow. It is designed to support reflection after incorrect responses and to reduce friction in mapping quiz items back to evidence in the learner’s own materials, while preserving appropriate constraints before answers are officially revealed.