AI Transparency

Crivo generates structured lesson plans from the constraints you provide: learning objectives, BNCC codes, available time, class size, materials, and pedagogical approach. The system also reviews existing plans, identifying strengths and suggesting improvements based on established pedagogical frameworks (Bloom, Backward Design, Rosenshine, Retrieval Practice, Cognitive Load Theory, 5E, Direct Instruction).

Crivo does not guarantee that a generated plan will work exactly as intended in any classroom. Every class is unique, with dynamics, pacing, and needs that no AI system can fully anticipate. The generated plan is an informed starting point, not a rigid script. The teacher is always responsible for the final decision to use, adapt, or discard any generated content.

The prompts sent to AI models (Anthropic Claude, OpenAI GPT, DeepSeek) include: the text of your planning request, the extracted structured constraints (class profile, time, materials, BNCC objectives), and the pedagogical references relevant to the stage. Not sent: your name, email, school name, city, or any student data. AI providers do not receive information that could identify you or your students.

No. We do not use teacher or school data to train or fine-tune AI models. We also do not authorize providers (Anthropic, OpenAI, DeepSeek) to use prompts or responses for model training. Your lessons, plans, and reviews remain yours.

BNCC codes are used as a structured reference for aligning learning objectives. Crivo maintains a local database of BNCC codes and descriptions (Elementary Education and Early Childhood Education). Alignment validation is done in two layers: (1) exact code matching via local lookup, and (2) semantic AI analysis only when necessary. No AI 'guesses' BNCC codes. The system always works from the codes you provide and the local database.

Every pipeline step requires your approval before moving forward. Constraint extraction shows what the system understood from your text; you review and adjust. Validation points out conflicts; you choose how to resolve them. Generation produces the plan in three phases; you approve each one. The final review shows suggestions from five pedagogical specialists; you accept or reject each suggestion individually. At no point does Crivo publish, share, or apply a plan without your explicit decision.

Hallucination is when AI generates incorrect or fabricated information that appears truthful. Crivo reduces this risk with overlapping layers of protection. The first is structured extraction: the system converts your free text into validated fields before generating any content. This reduces the space for the model to improvise. Next, the BNCC codes you provide are validated against our local database, a fixed source of truth, not against what the model thinks it knows. During generation, we force models to respond in validated JSON schema format, which significantly reduces divergence. Review then exposes the plan to five independent pedagogical specialists, it is unlikely the same error convinces all of them. The final layer is you: every section of the plan requires your approval before it can be used.

Before using any plan generated by Crivo in the classroom, we recommend checking five things. First: the specific objectives match what your class needs to learn at that moment, not before, not after. Second: the estimated time for each activity is realistic for your students; different classes have different pacing. Third: the listed materials are actually available at your school. Fourth: differentiation and accessibility strategies address your students' specific needs. Fifth: the proposed assessment is appropriate for the class profile. Use the plan as an informed foundation, adapt it with your knowledge of the class, and trust your professional judgment.