Rodolfo Azevedo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Rodolfo Azevedo has a total of 161 co-authors. Here are the top co-authors by number of shared publications:
G. Araujo
State University of Campinas (UNICAMP), Campinas, Brazil
11 shared publicationsGuido Araujo
11 shared publicationsSandro Rigo
Institute of Computing, University of Campinas, Campinas, Brazil
11 shared publications- Guido Araujo
Universidade Estadual de Campinas (UNICAMP)
8 shared publications Paulo Centoducatte
Institute of Computing - University of Campinas (UNICAMP) - Brazil
8 shared publications- Ricardo Caceffo
UNICAMP
7 shared publications
Latest Publications
- Open Access
- 22 August 2026
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
- Vol. 58 (15) , 1-34
Automatic Question Generation with Large Language Models: A Survey
Test-based learning is effective in fostering knowledge retention, but manually creating assessment questions remains time-consuming and limits personalized student practice. The advent of Large Language Models (LLMs) has introduced new possibilities for Automatic Question Generation (AQG). Motivated by this context, this survey provides a comprehensive overview of AQG using LLMs, focusing on educational applications. Following the PRISMA methodology, we reviewed 132 studies published between 2023 and 2025. Our contributions include a taxonomy of question types by response openness, an analysis of AQG efforts across knowledge fields, educational levels, evaluation strategies, and difficulty control. We also identify recurring challenges and research opportunities.
- 22 August 2026
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
- Vol. 58 (15) , 1-34
Open AccessAutomatic Question Generation with Large Language Models: A Survey
Test-based learning is effective in fostering knowledge retention, but manually creating assessment questions remains time-consuming and limits personalized student practice. The advent of Large Language Models (LLMs) has introduced new possibilities for Automatic Question Generation (AQG). Motivated by this context, this survey provides a comprehensive overview of AQG using LLMs, focusing on educational applications. Following the PRISMA methodology, we reviewed 132 studies published between 2023 and 2025. Our contributions include a taxonomy of question types by response openness, an analysis of AQG efforts across knowledge fields, educational levels, evaluation strategies, and difficulty control. We also identify recurring challenges and research opportunities.
- Open Access
- 9 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 88219-88236
Generative AI in Introductory Programming for Non-CS STEM Students: A Two-Semester Study
The emergence of generative artificial intelligence tools such as ChatGPT and GitHub Copilot has introduced new possibilities for programming education. However, few published works have described how these tools have been effectively integrated into undergraduate courses, particularly those with analyses grounded in full semesters of classroom experience. This paper addresses this gap by documenting and examining the use of generative AI in an introductory programming course for non-CS STEM undergrad-uate students. The course was redesigned to emphasize problem understanding, prompt formulation, and evaluation of AI-generated code, rather than focusing solely on manual implementation. The redesign also allowed expanded curriculum coverage, including additional topics not addressed in the traditional offering, such as debugging and testing.We analyze data from two semesters in which generative AI tools (ChatGPT and GitHub Copilot) were incorporated and compare results with two semesters of the traditional course format. The results suggest that the AI-supported format maintained high student engagement in assignment submissions throughout the semester, even after students reached the minimum passing grade. Student feedback indicated increased motivation and perceived relevance of the tools, along with new challenges related to debugging and problem decomposition.We conclude with practical recommendations and lessons learned for course planning and assessment design when integrating generative AI into programming education.
- 9 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 88219-88236
Open AccessGenerative AI in Introductory Programming for Non-CS STEM Students: A Two-Semester Study
The emergence of generative artificial intelligence tools such as ChatGPT and GitHub Copilot has introduced new possibilities for programming education. However, few published works have described how these tools have been effectively integrated into undergraduate courses, particularly those with analyses grounded in full semesters of classroom experience. This paper addresses this gap by documenting and examining the use of generative AI in an introductory programming course for non-CS STEM undergrad-uate students. The course was redesigned to emphasize problem understanding, prompt formulation, and evaluation of AI-generated code, rather than focusing solely on manual implementation. The redesign also allowed expanded curriculum coverage, including additional topics not addressed in the traditional offering, such as debugging and testing.We analyze data from two semesters in which generative AI tools (ChatGPT and GitHub Copilot) were incorporated and compare results with two semesters of the traditional course format. The results suggest that the AI-supported format maintained high student engagement in assignment submissions throughout the semester, even after students reached the minimum passing grade. Student feedback indicated increased motivation and perceived relevance of the tools, along with new challenges related to debugging and problem decomposition.We conclude with practical recommendations and lessons learned for course planning and assessment design when integrating generative AI into programming education.
- Open Access
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Investigating the Impact of Automated Code Quality Feedback in an Embedded Systems Course
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-12
Open AccessInvestigating the Impact of Automated Code Quality Feedback in an Embedded Systems Course
- Open Access
- 28 October 2025
- Published by Sociedade Brasileira de Computação
- p. 181-192
Large-Scale RISC-V Processor Verification Using Automated Design Inspection and a Generic Simulation Method
In recent years, RISC-V ISA has gained popularity and research on RISC-V processor verification has increased. However, most studies focus on a small number of cores and rely on implementation-dependent characteristics. To address this limitation, we propose a generic verification framework that monitors high-level processor states through the fetch interface and the register file write port. The execution traces are compared against a reference model. Our framework leverages a Large Language Model to collect the necessary files and signals of new processors, allowing manual code reductions of 85%. We evaluated the framework by testing 21 LLMs in the inspection task and simulating a custom benchmark on 21 processors, identifying 16 bugs across 8 different cores.
- 28 October 2025
- Published by Sociedade Brasileira de Computação
- p. 181-192
Open AccessLarge-Scale RISC-V Processor Verification Using Automated Design Inspection and a Generic Simulation Method
In recent years, RISC-V ISA has gained popularity and research on RISC-V processor verification has increased. However, most studies focus on a small number of cores and rely on implementation-dependent characteristics. To address this limitation, we propose a generic verification framework that monitors high-level processor states through the fetch interface and the register file write port. The execution traces are compared against a reference model. Our framework leverages a Large Language Model to collect the necessary files and signals of new processors, allowing manual code reductions of 85%. We evaluated the framework by testing 21 LLMs in the inspection task and simulating a custom benchmark on 21 processors, identifying 16 bugs across 8 different cores.
- Open Access
- 28 October 2025
- Published by Sociedade Brasileira de Computação
- p. 434-445
Extração Eficiente de MFCCs em FPGA: Uma implementação Aberta e Flexível
Este artigo apresenta um conjunto de IPs de código aberto e parametrizáveis para extração de MFCCs em FPGAs, otimizado para aplicações de borda e de tempo real. A implementação em um FPGA Kintex-7 processa um quadro de áudio em apenas 56 µs, superando em 1.62 vezes um CPU de alto desempenho (AMD Ryzen 7 7700). A arquitetura é eficiente em recursos, utilizando cerca de 5900 LUTs e 31 DSPs, e mantém alta precisão (erro absoluto médio de 8.18%), validando-se como uma solução de alta performance e baixo consumo para análise de áudio em tempo real.
- 28 October 2025
- Published by Sociedade Brasileira de Computação
- p. 434-445
Open AccessExtração Eficiente de MFCCs em FPGA: Uma implementação Aberta e Flexível
Este artigo apresenta um conjunto de IPs de código aberto e parametrizáveis para extração de MFCCs em FPGAs, otimizado para aplicações de borda e de tempo real. A implementação em um FPGA Kintex-7 processa um quadro de áudio em apenas 56 µs, superando em 1.62 vezes um CPU de alto desempenho (AMD Ryzen 7 7700). A arquitetura é eficiente em recursos, utilizando cerca de 5900 LUTs e 31 DSPs, e mantém alta precisão (erro absoluto médio de 8.18%), validando-se como uma solução de alta performance e baixo consumo para análise de áudio em tempo real.
- Open Access
- 28 October 2025
- Published by Sociedade Brasileira de Computação
- p. 89-96
Processor CI Inspector: Detectando parâmetros automaticamente em processadores RISC-V
Este artigo apresenta uma ferramenta, a ser incorporada em um ambiente de integração contínua, que visa, automaticamente, detectar parâmetros de processadores RISC-V. Utilizando-se de softwares open-source como simuladores de linguagem de descrição de hardware (e.g. Icarus Verilog e Verilator) e frameworks de cossimulação (e.g. Cocotb) foi projetada uma aplicação escalável que auxilia na escolha entre diferentes implementações de CPUs. Resultados preliminares constataram a eficácia da ferramenta, realçando a capacidade da metodologia em identificar características essenciais à seleção de processadores em diferentes contextos. Foi possível detectar 6 parâmetros de configuração de 9 implementações abertas distintas com sucesso.
- 28 October 2025
- Published by Sociedade Brasileira de Computação
- p. 89-96
Open AccessProcessor CI Inspector: Detectando parâmetros automaticamente em processadores RISC-V
Este artigo apresenta uma ferramenta, a ser incorporada em um ambiente de integração contínua, que visa, automaticamente, detectar parâmetros de processadores RISC-V. Utilizando-se de softwares open-source como simuladores de linguagem de descrição de hardware (e.g. Icarus Verilog e Verilator) e frameworks de cossimulação (e.g. Cocotb) foi projetada uma aplicação escalável que auxilia na escolha entre diferentes implementações de CPUs. Resultados preliminares constataram a eficácia da ferramenta, realçando a capacidade da metodologia em identificar características essenciais à seleção de processadores em diferentes contextos. Foi possível detectar 6 parâmetros de configuração de 9 implementações abertas distintas com sucesso.
- Open Access
Who Introduces and Who Fixes? Analyzing Code Quality in Collaborative Student's Projects
This paper investigates code quality education by analyzing how errors are introduced and corrected in group projects within an embedded systems course. We identify who introduces errors, who fixes them, and when these actions occur. Students learn code quality rules for C and embedded systems. We address three questions: RQ1: What is the impact of group formation on code quality? RQ2: How do students interact to fix code issues? RQ3: When are issues introduced and resolved? We analyzed data from eight individual labs and two group projects involving 34 students. The course provides continuous, automated feedback on code quality. Findings show that the most active contributors often introduce the most issues. Many issues are fixed late in the project. Individual labs tend to have fewer issues due to their structured nature. Most problems are fixed by the original author, while cross-student fixes take longer, especially in shared code. Critical issues are fixed quickly, but non-critical ones may be ignored, showing a focus on functionality over quality.
Open AccessWho Introduces and Who Fixes? Analyzing Code Quality in Collaborative Student's Projects
This paper investigates code quality education by analyzing how errors are introduced and corrected in group projects within an embedded systems course. We identify who introduces errors, who fixes them, and when these actions occur. Students learn code quality rules for C and embedded systems. We address three questions: RQ1: What is the impact of group formation on code quality? RQ2: How do students interact to fix code issues? RQ3: When are issues introduced and resolved? We analyzed data from eight individual labs and two group projects involving 34 students. The course provides continuous, automated feedback on code quality. Findings show that the most active contributors often introduce the most issues. Many issues are fixed late in the project. Individual labs tend to have fewer issues due to their structured nature. Most problems are fixed by the original author, while cross-student fixes take longer, especially in shared code. Critical issues are fixed quickly, but non-critical ones may be ignored, showing a focus on functionality over quality.
- Open Access
- 7 April 2025
- Published by Sociedade Brasileira de Computação
- p. 91-95
Além da Corretude: Investigando os Limites da Correção Automática ao Analisar Códigos Corretos
Este artigo não possui um resumo.
- 7 April 2025
- Published by Sociedade Brasileira de Computação
- p. 91-95
Open AccessAlém da Corretude: Investigando os Limites da Correção Automática ao Analisar Códigos Corretos
Este artigo não possui um resumo.
- Open Access
- 5 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 305-306
Micro-Specialization as a Solution to Open-Ended Project
This poster presents our experimental design, which will be implemented next semester, and shares insights from a preliminary version already executed. We will describe a method for managing the diverse content requirements of groups engaged in advanced open-ended projects. Our approach, termed "micro-specialization," involves students selecting specific tracks to acquire the knowledge needed for their chosen project. These tracks are comprised of individual activities designed to teach new knowledge that hasn't been covered before. Each student specializes in one track, gaining expertise, and then brings this new knowledge back to their group. The concept of individuals within a group specializing in different areas during an open-ended project is not extensively explored in the literature. While collaborative learning has been shown to benefit students in such projects, and flexible course designs with personalized objectives enhance motivation and academic performance, traditional methodologies like Jigsaw are typically applied to brief, specific activities under close faculty guidance. In contrast, our approach brings students together with the shared objective of completing an open-ended project, requiring them to collaboratively manage and execute the necessary tasks. This method introduces a novel application of collaborative strategies in a less structured educational environment. We plan to assess the effectiveness of this methodology by analyzing students' perceptions of their knowledge before and after completing their projects. Through a evaluation at the end of the semester, we aim to measure students' knowledge in all specializations to verify if collaborative learning occurs. Additionally, we will analyze individual student contributions to the project code. Using git blame, we can associate each section of modified code with a student and their track specialization. This will allow us to measure how much each student is contributing to their specialization and their ability to contribute to other specializations.
- 5 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 305-306
Open AccessMicro-Specialization as a Solution to Open-Ended Project
This poster presents our experimental design, which will be implemented next semester, and shares insights from a preliminary version already executed. We will describe a method for managing the diverse content requirements of groups engaged in advanced open-ended projects. Our approach, termed "micro-specialization," involves students selecting specific tracks to acquire the knowledge needed for their chosen project. These tracks are comprised of individual activities designed to teach new knowledge that hasn't been covered before. Each student specializes in one track, gaining expertise, and then brings this new knowledge back to their group. The concept of individuals within a group specializing in different areas during an open-ended project is not extensively explored in the literature. While collaborative learning has been shown to benefit students in such projects, and flexible course designs with personalized objectives enhance motivation and academic performance, traditional methodologies like Jigsaw are typically applied to brief, specific activities under close faculty guidance. In contrast, our approach brings students together with the shared objective of completing an open-ended project, requiring them to collaboratively manage and execute the necessary tasks. This method introduces a novel application of collaborative strategies in a less structured educational environment. We plan to assess the effectiveness of this methodology by analyzing students' perceptions of their knowledge before and after completing their projects. Through a evaluation at the end of the semester, we aim to measure students' knowledge in all specializations to verify if collaborative learning occurs. Additionally, we will analyze individual student contributions to the project code. Using git blame, we can associate each section of modified code with a student and their track specialization. This will allow us to measure how much each student is contributing to their specialization and their ability to contribute to other specializations.
- Open Access
- 23 October 2024
- Published by Sociedade Brasileira de Computação
- p. 1-8
ProcessorCI: Integração Contínua para processadores RISC-V em FPGAs
Este artigo apresenta uma infraestrutura automatizada para a verificação de processadores RISC-V, utilizando técnicas de integração contínua combinadas com FPGAs de múltiplos fornecedores, como Altera/Intel, Xilinx/AMD, Gowin e Lattice. Foi projetado um ambiente robusto e escalável que facilite a detecção precoce de falhas em diversas implementações de processadores RISC-V, garantindo a conformidade com as especificações da ISA. Resultados preliminares demonstram a eficácia da abordagem, destacando a capacidade da metodologia em identificar rapidamente erros e coletar dados essenciais para a seleção e verificação de processadores.
- 23 October 2024
- Published by Sociedade Brasileira de Computação
- p. 1-8
Open AccessProcessorCI: Integração Contínua para processadores RISC-V em FPGAs
Este artigo apresenta uma infraestrutura automatizada para a verificação de processadores RISC-V, utilizando técnicas de integração contínua combinadas com FPGAs de múltiplos fornecedores, como Altera/Intel, Xilinx/AMD, Gowin e Lattice. Foi projetado um ambiente robusto e escalável que facilite a detecção precoce de falhas em diversas implementações de processadores RISC-V, garantindo a conformidade com as especificações da ISA. Resultados preliminares demonstram a eficácia da abordagem, destacando a capacidade da metodologia em identificar rapidamente erros e coletar dados essenciais para a seleção e verificação de processadores.