Little Scientist: architecture and implementation of a controlled artificial intelligence agent for scientific and technological literacy in basic education
Year: 2026
Type: article
Abstract: The objective of this study is to analyze the development and implementation of a controlled artificial intelligence agent, named Little Scientist (Cientista Mirim), aimed at supporting K-12 students in the process of Scientific-Technological Literacy (AlfaCT). The research problem investigates how a supervised AI architecture restricted to a curated document corpus can foster critical thinking, investigative reasoning, and intellectual curiosity without incurring the epistemological risks of hallucination and plagiarism typical of general generative models. The methodology is exploratory with a qualitative approach, structured across four dimensions: definition of the pedagogical scope, computational architecture based on Retrieval-Augmented Generation (RAG), interface configuration in guided wizard and free modes with persona support, and rigorous response control tied to a vector database (sqlite-vec in SQLite, with 1536-dimensional vectors hosted on Azure). The results demonstrate that Little Scientist acts as an effective cognitive mediator, offering personalized support, transparency through prompt metacognition, and full source traceability via precise citation of files, pages, and lines. A key pedagogical differentiator is the agent's ability to explicitly state when information is absent from the corpus, modeling the scientific attitude of acknowledging knowledge limits. It is concluded that controlled AI agents represent a viable and ethically sound alternative to strengthen school inquiry, combining technological innovation and pedagogical responsibility under irreplaceable teacher mediation.
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: hybrid
OpenAlex ID: W7214239016
