Technology in School Management: Automation and Quality in the Teaching Plans of Senai São Paulo

Lucas Tadeu Monteiro Guedes Fernandes Salomão; Flávia Pierrotti De Castro

DOI: 10.22167/2675-6528-202603182

Article derived from a Course Conclusion Work (TCC), with content based on the student’s original work and adapted to the editorial format of the E&S Magazine with the support of the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege for textual synthesis and organization.

Summary

The development of pedagogical documents, such as Teaching Plans and assessment instruments, aligned with the SENAI Professional Education Methodology (MSEP), demands technical rigor and often results in an overload of work for the teaching staff. Faced with this challenge, the effectiveness of the “MSEP Assistant”, a virtual assistant based on Generative Artificial Intelligence (LLM), developed to optimize school planning, was analyzed. The objective was to evaluate its effectiveness in two complementary dimensions: operational efficiency, measured by the reduction in planning time, and pedagogical quality, validated through thematic cycles that analyzed curricular fidelity and system usability. The applied research was conducted in two distinct phases: a quantitative efficiency diagnosis and a qualitative validation structured in four thematic cycles, using Google Classroom. The results of the first phase indicated a reduction of up to 90% in document preparation time. In the second phase, it was observed that the tool ensured methodological standardization, with 100% fidelity, and improved the quality of evaluation criteria, although it required human review for workload calculations. It was concluded that the integration of Artificial Intelligence in school management promoted robust operational efficiency and allowed teachers to refocus on pedagogical strategies and human knowledge mediation.

Keywords: Educational Efficiency; Generative Artificial Intelligence; SENAI Methodology for Professional Education; Pedagogical Planning.

1. Introduction

Professional Education requires rigorous alignment between the demands of the labor market and pedagogical practice in the classroom. In this context, the SENAI Methodology for Professional Education (MSEP) establishes precise guidelines for teaching planning, structuring teaching through the development of professional competencies in challenging and contextualized learning situations, focusing on real work scenarios (SENAI, 2019).

The elaboration of pedagogical documents, such as Teaching Plans and assessment instruments, demands technical rigor and significantly impacts the time management of the teaching staff. The inherent technical complexity of these processes often results in an overload of work. The traditional organizational model, often focused on excessive control bureaucracy, tends to prioritize the fulfillment of forms, which can diminish the educator’s autonomy and creativity (Libâneo, 2001).

The time dedicated to methodological formatting directly competes with the time for pedagogical strategy, generating a bottleneck in school management. This intensification of teaching work, characterized by the multiplicity of administrative tasks and time compression, is pointed out in sociological and occupational health literature as one of the main vectors of malaise and deprofessionalization (Nóvoa, 1992). Continuous overload acts as a chronic stressor, contributing to the development of Burnout Syndrome in the category (Carlotto, 2002).

In contrast, contemporary school management seeks to overcome the fragmented administrative vision, mobilizing the team and technology to optimize processes and ensure teaching quality (Lück, 2000). The advancement of Generative Artificial Intelligence (LLM) opens new possibilities for supporting teaching work, acting as a versatile and efficient pedagogical support in automating repetitive tasks (Vogel, 2025).

This work presents the case study of a virtual assistant, named “MSEP Assistant”, specifically developed to help teachers in the construction of documents aligned with MSEP. The tool guides the teacher in selecting technical and socio-emotional skills, suggesting industrial contexts, and structuring the evaluation criteria, functioning as technological support for pedagogical coordination.

The justification for this research lies in the pressing need to modernize pedagogical management practices, seeking tools that not only standardize processes but also promote occupational health and the efficiency of the school team. By mitigating excessive bureaucracy, one of the pathogenic factors of teacher stress (Carlotto, 2002), the introduction of virtual assistants aims to give teachers back time to focus on knowledge mediation and human relationships, which are essential and irreplaceable (Vogel, 2025). Therefore, the objective of this work is to analyze the effectiveness of using an artificial intelligence-based virtual assistant in optimizing teacher planning, evaluating its effectiveness in two complementary dimensions: operational efficiency, measured by the reduction in planning time (Phase 1), and pedagogical quality, validated through thematic cycles that analyzed everything from curricular fidelity to system usability (Phase 2).

2. Material and Methods

The present research was characterized as a case study, an essential method for investigations that require an in-depth analysis of contemporary phenomena embedded in their real-life context (Yin, 2015). Of an applied nature, the investigation was based on the principles of Design Science Research (DSR), a methodological approach aimed at solving practical organizational problems through the development, iterative testing, and validation of innovative technological artifacts (Wieringa, 2014).

To ensure robustness in the evaluation of the artifact’s operational and pedagogical efficiency, a mixed-methods (quantitative-qualitative) research approach was adopted. This choice was based on the premise that the systematic association and cross-referencing of quantitative data with qualitative narrative analyses provide a superior and complementary methodological understanding of the object of study (Creswell, 2010).

The study was conducted in several vocational training units located in the state of São Paulo, which serve approximately 73,000 students in the Industrial Apprenticeship and Technical Courses segments. Twenty-eight teachers from the institution participated in the study, selected by voluntary adhesion and covering diverse technological areas.

Of these participants, 11 integrated Phase 1, focused on efficiency diagnosis, and 17 participated in Phase 2, dedicated to qualitative validation in cycles. All participants were informed about the research objectives, ensuring the anonymity and confidentiality of individual responses, in accordance with ethical guidelines for research involving human beings.

For the experiment, a virtual assistant, named “MSEP Assistant”, was developed to specifically help teachers in the elaboration of teaching plans according to the SENAI Methodology of Professional Education (MSEP). The tool guides the teacher in selecting technical and socio-emotional skills, suggesting industrial contexts, and structuring the evaluation criteria.

The virtual assistant’s architecture was not based on training its own model, but on the technique of grounding. A general-purpose Large Language Model (LLM), Google Gemini, was conditioned, at each interaction, by a specific documentary base for the generation of Teaching Plans and Learning Situations.

This documentary base included the full document of the SENAI Methodology for Professional Education (SENAI, 2019), incorporated into the model’s system instructions as a permanent normative reference. Additionally, the Course Plan of the curricular unit selected by the instructor constituted the primary source of curricular data, being provided in PDF format and converted into structured text.

Data collection was structured in a Virtual Learning Environment, Google Classroom, where the experimental design was organized in two distinct phases, allowing for an evolutionary analysis of the tool. This approach aimed to evaluate the effectiveness of the virtual assistant in two complementary dimensions.

Phase 1, named Efficiency Diagnosis, occurred between March and April 2025. This exploratory stage focused on quantitative measurement, comparing the average time for manual creation of pedagogical documents versus AI-assisted creation, with the objective of identifying the main operational bottlenecks of the traditional process.

Phase 2, Qualitative Validation in Cycles, was carried out between November and December 2025. In this deepening stage, the tool, already adjusted after Phase 1, was submitted to four thematic validation cycles, also conducted via Google Classroom, to analyze the pedagogical quality of the generated content.

Cycle 1 of Phase 2 evaluated fidelity to the Course Plan, verifying the assistant’s ability to correctly select the Technical, Social, and Organizational Capabilities provided for in the chosen course plan. Cycle 2 focused on the quality of the Learning Situation, analyzing whether the scenario generated by the AI presented adequate contextualization and whether the proposed Challenging Learning Strategy was feasible and relevant to the professional level.

Cycle 3 verified the tool’s capacity to unfold capabilities into measurable and objective evaluation indicators, facilitating the construction of assessment instruments. Finally, Cycle 4 evaluated the user experience (UX) and the impact on reducing teacher stress, measuring the ease of use of the interface and the perception of decreased time spent on bureaucratic tasks.

Data collection occurred through electronic forms applied at the end of each cycle. Each cycle, with its respective form, was structured and published on Google Classroom. For quantitative analysis, the Likert scale (from 1 to 5 or 1 to 4) was used to measure the degree of agreement and satisfaction of the faculty.

The quantitative data were tabulated and processed for the generation of descriptive statistics. For the qualitative analysis, responses in open fields were considered, which were submitted to content analysis (Bardin, 2016) to identify patterns, suggestions for improvement, and perceptions about the impact of technology on the teaching routine.

3. Results and Discussion

The research results were structured in two complementary phases, reflecting the methodological design adopted to evaluate the effectiveness of the virtual assistant. The first phase, quantitative in nature, focused on diagnosing operational efficiency and the initial acceptance of the tool by teachers. The second phase, qualitative in nature, delved into the pedagogical validation of the artifact, exploring curricular fidelity, the quality of learning situations, the accuracy of evaluation criteria, and the usability of the system, according to the established thematic cycles.

Initial Diagnosis (Phase 1 – Mar to Apr/2025)

The initial validation phase involved eleven teachers, all with a high level of proficiency in the SENAI Methodology for Professional Education (MSEP). This selection of participants was strategic to ensure a critical and in-depth analysis of the pedagogical fidelity of the content generated by the virtual assistant. The teachers’ experience allowed for a robust evaluation of the tool, identifying both its strengths and areas that required improvement, contributing to the refinement of the instruction models (prompts) used by artificial intelligence.

The quantitative results of this phase evidenced a high adherence of the tool to the SENAI methodology requirements. A notable assertiveness was observed in the suggestion of Evaluation Criteria, with an average of 4.55 on a scale of 1 to 5. Similarly, the integration of Socioemotional Capabilities also achieved an average of 4.55, indicating that the tool managed to optimize historically complex aspects in the manual preparation of pedagogical documents. Technical Capabilities also presented a high average of 4.45, exceeding the quality target of 4.00 established for the project.

The clarity of the proposed challenge obtained an average of 4.27, and the overall clarity of the result generated by the tool was evaluated with an average of 4.0. However, the initial diagnosis revealed a specific improvement opportunity in the definition of the Expected Challenge Outcome, which obtained an average of 4.00. The professors pointed out that, although the text generated by artificial intelligence was correct, it lacked greater specificity regarding the technical deliverables. This feedback was fundamental for refining the prompts, directing adjustments for the second round of tests and ensuring greater precision.

In terms of user acceptance and satisfaction, the solution achieved a high Net Promoter Score (NPS). Ten out of eleven instructors gave a score of ten, and one instructor gave a score of nine, indicating that, even in the pilot phase, the tool was perceived as a significant added value to the teaching routine. This high recommendation rate suggests that the virtual assistant has great potential for widespread adoption and integration into pedagogical planning practices, facilitating work and promoting efficiency in school management.

The detailed analysis of the collected data allowed for the identification of specific bottlenecks in the manual planning process. The “Final Consolidation” stage, which involves document formatting and adaptation to technical standards, consumed an average of 250 minutes in the traditional process. With the use of the Virtual Assistant, this stage was drastically reduced to just 16 minutes, as the tool generates the standardized document directly, requiring only final review from the instructor. This reduction represents a time saving of approximately 93.6% in this task.

Another significant gain in efficiency was observed in the “Development of the Learning Situation” stage. The time dedicated to this task, which was previously 46 minutes for creation from scratch, was reduced to 10 minutes with the use of the assistant, which generates suggestions for validation. This 78.3% decrease in development time demonstrates the tool’s ability to eliminate operational work of typing and formatting, freeing up the instructor to focus on pedagogical validation and teaching strategy, as pointed out by Vogel (2025) regarding the potential of AI in supporting teaching work.

The decomposition of time by task revealed that the total average manual planning time was 359 minutes, while with the artificial intelligence assistant, this time was reduced to 43 minutes. This reduction of approximately 88% in the total time for preparing pedagogical documents reinforces the robustness of the tool in optimizing operational efficiency. The data confirm that the technology is especially effective in automating repetitive tasks, allowing educators to dedicate more time to activities of greater pedagogical and strategic value.

A cross-analysis between the level of experience of the faculty and the satisfaction index (NPS) revealed that the tool obtained maximum approval (NPS 10.0) among “MSEP Specialists”, i.e., faculty with complete training in the methodology. This finding suggests that the virtual assistant not only simplifies the work but also respects the technical rigor required by the most qualified professionals, which is crucial for the acceptance and integration of technology in the educational environment. Additionally, the technical robustness indicator showed that in 100% of interactions, the artificial intelligence faithfully reproduced the Course Plan Capabilities, without generating false data or “hallucinations”, ensuring the legal and pedagogical security of the document.

Final diagnosis (Phase 2 – Nov to Dec/2025)

Fidelity and Coherence of Data Extraction (Cycle 1)

The first cycle of the second phase, with the participation of seventeen faculty members, focused on the fidelity and coherence of data extraction. The main objective was to ensure that artificial intelligence correctly interpreted the Course Plan (PC), without distorting or altering mandatory curricular data. The results demonstrated that the tool achieved a level of excellence in capturing structured data, confirming AI’s ability to perform the “bureaucratic reading” of the curriculum with superior accuracy to humans, eliminating copy errors and ensuring methodological standardization.

In 100% of cases, artificial intelligence correctly identified the Curricular Units (CUs) and Technical and Socio-emotional Skills, with only minor notes on the visual ordering of the list, but without compromising the content. The fidelity of the Course Information, such as Course Name, Workload, and Dates, was also 100%, with small requests for formatting adjustments. These results validate the hypothesis that AI can optimize repetitive administrative tasks, freeing up faculty to focus on more strategic aspects of teaching, according to Lück’s (2000) perspective on contemporary school management.

Quality of the Learning Situation (Cycle 2)

The biggest challenge for artificial intelligence was not limited to data copying, but rather to its integration into a cohesive and contextualized pedagogical narrative. In the evaluation of the creation of Learning Situations (SA), 70% of the generated SAs (twelve out of seventeen) presented an “Exemplar” or “Good” integration, where socio-emotional skills were organically inserted into the technical challenge. This indicates a significant capacity of the tool to contextualize teaching, a pillar of the SENAI Methodology (SENAI, 2019).

However, 30% of the integrations were considered “Forced” or “Theoretical”, where artificial intelligence cited the capability (e.g., “Teamwork”) but failed to create a corresponding practical activity. This limitation reinforces the need for faculty curation to ensure pedagogical depth and applicability. Open-ended comments from faculty revealed critical perceptions of the AI’s behavior, highlighting the importance of the “Prompt Role”, where the quality of the teacher’s initial description directly impacts the quality of the generated plan, confirming that the tool empowers the teacher but does not replace their pedagogical intentionality.

Errors in “Mathematical Logic” were identified, such as inverted schedules (class from 10 PM to 8 PM) and incorrect summation of workload, indicating that generative artificial intelligence still presents difficulties with exact calculations. The “Chocolate Cake Test,” where the AI obeyed a request to include a cake recipe in an electricity plan, demonstrated the absence of “merit judgment” or “guardrails” against absurd content. This finding is crucial for ethical discussion, as it highlights that, although efficient, AI requires mandatory human review to ensure the pedagogical relevance and safety of the generated content, an aspect that Vogel (2025) points out as a limitation of LLMs.

Generated Learning Situations Assessments

Cycle 2 evaluated the creative capacity of artificial intelligence in generating contextualized Learning Situations (LS), focusing on three pillars of the MSEP: fidelity to the requested teaching strategy, quality of the industrial context, and clarity of the final product (deliverable). The tool obtained averages above 4.0 in all aspects, exceeding the quality target established for the Minimum Viable Product (MVP). The clarity of the Expected Outcome was the strongest point, with an average of 4.30, where faculty reported that the tool is extremely precise in defining what the student should deliver, facilitating subsequent evaluation.

The quality of the industrial context obtained an average of 4.15, demonstrating the high capacity of artificial intelligence in creating engaging and realistic narratives, with a professor citing that “the text places the student within the problem”. Fidelity to the pedagogical strategy, although positive, obtained the lowest score, with an average of 4.09, indicating a slight difficulty of AI in differentiating nuances of more complex strategies. This suggests that, for more refined pedagogical approaches, such as the Integrated Project, the professor’s curation is still indispensable to ensure the rigor of the format and the necessary depth.

The analysis of open-ended comments allowed for the identification of specific limitations in certain pedagogical strategies. There was confusion between “Project” and “Problem Situation,” where the AI generated a challenge that more closely resembled an extended problem situation, without the progressive steps typical of a project. In “Applied Research,” the generated content was considered superficial and failed to integrate some selected technical capabilities. Furthermore, the AI demonstrated “overzealousness” in one instance, suggesting a “Slide Presentation” as an extra deliverable, even without a selected oral communication capability, indicating that the tool sometimes tries to “gild the lily” beyond what was requested.

Evaluation Criteria and Planning (Cycle 3)

The third cycle evaluated the tool’s ability to structure the Evaluation Criteria, fundamental for the MSEP, and maintain the internal coherence of the Teaching Plan, including the distribution of workload and content. The generation of Evaluation Criteria was the highlight of this cycle, with an average of 4.38. The tool demonstrated high competence in breaking down capabilities (e.g., “Analyze…”) into observable and measurable criteria, a task that is often challenging for novice instructors. The adequacy of the evaluation model was also positive, with an average of 4.00, indicating that the AI correctly chose between dichotomous evaluation (Yes/No) for simple tasks and gradual (Rubrics) for complex socio-emotional competencies.

However, the Workload and Schedule indicator was the only one to remain in the critical zone, with an average of 2.92. Most failures involved mathematical inconsistencies, such as summation errors, where the AI frequently generated plans with a total workload lower or higher than that available in the Curricular Unit. There was also disregard for the calendar, with the tool ignoring end dates or academic days in several tests. Furthermore, instructors reported text truncation in long tables, where the AI tended to summarize texts using ellipses, which is inadequate for an official document.

These results confirm a known limitation of Large Language Models (LLMs), which are probabilistic text models and not logical calculators. For this study, this finding is valuable, as it proves that the tool acts as a “Creative Assistant”, but does not eliminate the need for human review for logical and chronological data validation. The need for human curation to correct these mathematical and formatting errors is a crucial point that delimits the role of technology, reinforcing that artificial intelligence is a support, but not a substitute for the educator’s judgment and expertise.

Usability Evaluation (Cycle 4)

The last validation cycle focused on human-computer interaction, evaluating process fluidity, interface aesthetics, and system robustness against errors. The interface design was widely praised, with an average of 4.50, being considered clean, professional, and intuitive by users. However, the workflow obtained an average of 3.00, indicating usability issues. The main friction point was the lack of system feedback, where buttons became “disabled” without explaining to the user which mandatory field was missing, generating frustration and compromising the user experience.

Regarding stability, the system obtained an average of 3.60, proving robust in the sense of “not crashing”, but it failed in logical data validation. Significant bugs were found, such as the ability of teachers to generate plans with retroactive dates (last year) and impossible schedules (starting at 10 PM and ending at 8 PM). The system accepted this data without issuing alerts, which compromises reliability if there is no human review. The export functionality to .docx obtained an average of 3.70 and was considered functional, but with reservations, as the formatting of complex tables sometimes “broke” when attempting fine adjustments. A recurring suggestion was the automatic inclusion of the sum of hours for the Learning Situation, which currently needs to be calculated manually.

In summary, the research demonstrated that the virtual assistant based on artificial intelligence is an effective tool in optimizing teaching planning, proving its ability to drastically reduce the time dedicated to bureaucratic and formatting tasks. On the pedagogical dimension, the tool ensured total adherence to the MSEP curricular regulations and elevated the technical level in structuring classes and evaluation criteria, generating clear and contextualized proposals. However, the inconsistencies identified in the workload and schedule logic, as well as the need for human curation to ensure merit judgment and correction of mathematical errors, delimit the role of technology, reinforcing that artificial intelligence acts as advanced support, but does not replace the pedagogical intentionality and human mediation of the educator.

4. Conclusion

This study aimed to analyze the effectiveness of using an artificial intelligence-based virtual assistant in optimizing teaching planning, evaluating its operational efficiency and pedagogical quality. A robust optimization of operational efficiency was verified, with the tool significantly reducing the time dedicated to the elaboration of pedagogical documents. A decrease of up to 88% in total planning time was observed, freeing up teachers from bureaucratic and formatting tasks. In the dimension of pedagogical quality, it was identified that the assistant ensured 100% fidelity to the curricular regulations of the SENAI Methodology for Professional Education, raising the technical level in structuring classes and defining clear and contextualized evaluation criteria. The main contribution lies in the technology’s ability to absorb exhaustive work, returning the focus to educators on pedagogical intentionality, human knowledge mediation, and teaching strategy, which positively impacts occupational health and school management efficiency.

However, the study revealed important limitations that delimit the role of artificial intelligence. Inconsistencies in mathematical logic were identified, such as errors in summing workload hours and disregard for schedules, in addition to the absence of merit judgment in certain contexts, which required mandatory human review. Challenges in system usability were also observed, especially in the lack of clear user feedback and in the formatting of complex documents. For future studies and tool improvements, it is suggested to implement more robust logical validations for workload and schedule calculations, improve the workflow with interactive feedback, and include functionalities such as automatic hour summation. These findings reinforce that artificial intelligence acts as advanced and creative support, but does not replace the expertise, sensitivity, and human curation of the educator, which remain essential for ensuring pedagogical relevance and safety.

Bibliographic References

Bardin, L. 2016. Análise de conteúdo. Edições 70, São Paulo, SP, Brasil.

Carlotto, M.S. 2002. A síndrome de burnout e o trabalho docente. Psicologia em Estudo 7(1): 21-29.

Creswell, J.W. 2010. Projeto de pesquisa: métodos qualitativo, quantitativo e misto. 3ed. Artmed, Porto Alegre, RS, Brasil.

Libâneo, J.C. 2001. Organização e Gestão da Escola: teoria e prática. 4ed. Alternativa, Goiânia, GO, Brasil.

Lück, H. 2000. Perspectivas da gestão escolar e implicações quanto à formação de seus gestores. Em Aberto 17(72): 11-33.

Nóvoa, A. 1992. Os professores e as histórias da sua vida. In: Nóvoa, A. Vidas de professores. 2ed. Porto Editora, Porto, Portugal.

Serviço Nacional de Aprendizagem Industrial. 2019. Metodologia SENAI de educação profissional. SENAI/DN, Brasília, DF, Brasil.

Vogel, D. 2025. Transformando a educação com Large Language Models (LLMs): benefícios, limitações e perspectivas. Revista Caderno Pedagógico 22(4): 1-20.

Wieringa, R.J. 2014. Design science methodology for information systems and software engineering. Springer, Berlin, Alemanha.

Yin, R.K. 2015. Estudo de caso: planejamento e métodos. 5ed. Bookman, Porto Alegre, RS, Brasil.

Article originating from the Course Completion Work of the Specialization in School Management of the MBA USP/Esalq

To learn more about the course, click here and access the MBX Academy platform