Frequently Asked Questions
Answers to common questions about Blueprint.
How we ensure 100% Safety and Academic Rigor
To maintain the trust of educators, our AI agent operates within a Closed-Loop Knowledge Base. This means it does not "guess" what a 7-year-old should know; instead, it uses the provided papers as a structural and linguistic anchor.
1. Linguistic Guardrails (The Vocabulary Filter)
The AI analyzes the "word density" of the provided assessments. For example:
- Constraint: Instead of using "enormous" or "colossal," the AI is instructed to use "very big," matching the level found in the 2021 Revision Paper.
- Safety: This prevents students from being tested on words they haven't been taught in the MOE syllabus yet.
2. Structural Pattern Recognition
Our database allows the AI to recognize that a Primary 1 English Assessment must follow a specific "scaffolding":
- Grammar MCQs to build confidence (Tenses/Pronouns).
- Vocabulary MCQs to test world knowledge (Animal sounds/Transport).
- Synthesis & Cloze to test sentence logic and context.
- Comprehension to assess literal and inferential reading.
3. Verification through Verbatim Grounding
By including the 2018, 2019, and 2021 papers verbatim, we provide a "Source of Truth." If a teacher asks, "Why did the AI generate a question about past tense?" we can point directly to Paper 1 (2021) or Paper 4 (2018) where "yesterday" is used to trigger the past tense response "showed" or "ate".
The "Safe Generation" Workflow
- Input: Verbatim PDF data (The "Truth").
- Analysis: Extracting the "Pattern" (The "Rules").
- Generation: Creating new questions that fit the "Rules" but use new "Themes."
- Verification: Comparing the new question against the "Truth" to ensure it isn't too difficult.