i3 lab

International Summer School on LLM-based Agents for Software Engineering

Tue, 16 Sep 2025 00:00:00 +0000

From September 1st to 3rd, 2025, i3Lab (Quercus Software Engineering Group and INTIA, Universidad de Extremadura) successfully hosted the 1st International Summer School on LLM-based Agents for Software Engineering (LLMA4SE). The event brought together researchers, practitioners, and students from academia and industry to explore the fast-evolving field of LLM-based multi-agent systems and their applications in software engineering.

The opening day included two interesting talks:

AgentWare in Software Engineering: A Taxonomy of LLM-Based Multi-Agent Systems by Davide Di Ruscio, which led to stimulating discussions.
Generative AI Models Running in Your Own Infrastructure by Jesús M. González Barahona, who offered a deep perspective on openness in AI models.

Participants also enjoyed a hands-on workshop on TDD with the Help of AI, led by Jesús M. González Barahona and Giovanni Rosa. The day concluded with a cultural highlight: a guided tour of the UNESCO World Heritage city of Cáceres.

The second day offered a strong mix of research and practical applications. The day began with three particularly engaging presentations:

Multi-agents for Code Generation by Jie Zhang, addressing current challenges and approaches.
AI and Software Development: An Industry Paradigm Shift by Mario Corchero Jiménez and David Naranjo, presenting the industry perspective.
From Workflow-based to Fully Agentic Applications: smolagents and LangGraph by Antonio García Domínguez, who introduced agent development frameworks.

In the afternoon, Antonio García Domínguez guided participants through a highly practical workshop on developing agentic applications with human-in-the-loop via LangGraph. The day also included two enriching social activities: a guided visit to the Helga de Alvear Museum of Contemporary Art and a social dinner at Gran Hotel Don Manuel.

The final day, we enjoyed some thought-provoking sessions:

Olivier Perard on AI in Enterprise Architecture.
Marc Alier on Safe AI in Education - The LAMB Project, focusing on history, ethics, and risks.
Merve Astekin on Software Maintenance with Agentic LLMs, showcasing sustainable AI projects.

The summer school concluded with an advanced workshop, Multi-Agent LLMs and Knowledge Graphs for Monitoring and Sustaining Software Systems, delivered by Adela N. Videsjorden and Simeon Andersen Tverdal.

With the success of this first edition still fresh in our minds, we are already hard at work planning the next one. We fully expect the second edition of our school to be just as attractive and insightful as this inaugural event. Our goal is to maintain the high quality of content and the variety of the discussions to create a great meeting point for the community.

Stay tuned for future announcements!

MetrikaBox: An open framework for experimenting with audio classification

Thu, 28 Aug 2025 00:00:00 +0000

We open the (Metrika)Box

Thu, 28 Aug 2025 00:00:00 +0000

After months in the making, instead of keeping the information locked away like in a mystery box, we’ve finally opened it! Our lab is proud to share with you the distilled essence of our work: “MetrikaBox: An open framework for experimenting with audio classification”. The title says it all—no further notes needed.

With this publication, Jorge Perianez-Pascual’s PhD journey reaches a turning point, as it brings together years of research in a single paper. As described there, MetrikaBox’s impact is significant: it has supported the development of many papers and experiments, and it lies at the core of MetrikaMedia, a UEx spin-off company that defines itself as a SaaS solution for multimedia content measurement. And now, it’s available for everyone to explore! What will you do with it? Read the paper, play with the code—and don’t forget to tell us about it!

MetrikaBox is one of the main outcomes of musicgenia, a project funded by Grant CPP2021-008491 from MICIU/AEI/10.13039/50100011033 and by the European Union through NextGenerationEU/PRTR. The goal of musicgenia is to develop a cloud-based platform that provides AI-generated production music as a service for content creators and media, both online (live music generation) and offline (pre-recorded music generation). The platform brings clear benefits: (1) royalty-free music, (2) fully original compositions, (3) an easy way to find music that fits your content, and (4) flexible streaming options, where you pay per second instead of per song.

AI-Based System for Assistance in Minimally Invasive Renal Procedures Using Mixed Reality. First Steps

Tue, 29 Jul 2025 00:00:00 +0000

Everything Everywhere All at Once... with a single prompt

Mon, 14 Apr 2025 00:00:00 +0000

Following the direction set in one of our previous papers, we’re excited to share that our new article, “Prompt Once, Segment Everything: Leveraging SAM 2’s Potential for Infinite Medical Image Segmentation with a Single Prompt,” has just been published. Both papers are driven by a common question: Given the enormous resources typically required to develop specialized artificial intelligence models for a single task, could foundation models be used instead—without the need for costly optimization? To explore this, we focus on a task that is both technically challenging and critically important: medical image segmentation, where accurate results can directly impact patient care and save lives.

In this paper, we introduce a novel approach that harnesses the video segmentation capabilities of SAM 2 to reduce the number of prompts needed to segment an entire volume of medical images. The study begins by comparing the performance of SAM and SAM 2 in medical image segmentation. We then present our new method, which builds on SAM 2’s strengths. The results demonstrate that SAM 2 achieves an average improvement of 1.76 % in Jaccard Index and 1.49 % in Dice Score over SAM.

Our next objective is to develop a methodology that helps define the limits of what can be achieved using these foundation models.

We’re sure we’re missing something—after all, we can’t be everywhere at once. 😉 Got ideas? Come and see us, and let’s talk.

Prompt Once, Segment Everything is one of the outcomes of musicgenia, a project funded by the MICIU/AEI (grant CPP2021-008491, DOI: 10.13039/50100011033) and supported by the European Union through NextGenerationEU/PRTR.

Prompt Once, Segment Everything: Leveraging SAM 2 Potential for Infinite Medical Image Segmentation with a Single Prompt

Mon, 14 Apr 2025 00:00:00 +0000

Beyond Spectrograms: Rethinking Audio Classification from EnCodec's Latent Space

Sun, 16 Feb 2025 00:00:00 +0000

Spectrograms and Beyond!

Sun, 16 Feb 2025 00:00:00 +0000

It is a pleasure to start the year (even though we are already in February) by celebrating the publication of a new article from our laboratory. This time, it is “Beyond Spectrograms: Rethinking Audio Classification from EnCodec’s Latent Space”. In this work, we ponder the question: What if the latent representation of a neural audio codec were used in an audio classification pipeline? What we have discovered will surprise you. Take a look at the paper and let us know your thoughts.

This paper is one of the results of the thesis being developed by Jorge Perianez-Pascual, a member of our lab. Interestingly, the concept of using EnCodec’s latent space for classification was initially described as borderline unconventional, yet our results demonstrate its effectiveness. By stepping beyond traditional spectrogram-based methods, we uncover new possibilities for efficient and accurate audio classification.

This achievement would not have been possible without the collaboration of Álvaro Rubio Largo and Laura Escobar Encinas, two colleagues from the Universidad de Extremadura who contributed their multidisciplinary expertise both during the development and experimentation process and in the writing of the article.

Beyond Spectrograms is one of the results of musicgenia, a project funded by Grant CPP2021-008491 from MICIU/AEI/10.13039/50100011033 and by the European Union through NextGenerationEU/PRTR. The main goal of musicgenia is to develop a cloud-based platform that offers AI-generated production music as a service for content creators and media, both online (live music generation) and offline (pre-recorded music generation). The direct benefits of this platform include: (1) royalty-free music, (2) original music, (3) an easy way to find music that fits your content, and (4) streaming music, with a flexible consumption model where you pay per second rather than per song.

i3lab at the Winter School on AI and SE in Doha, Qatar

Sun, 19 Jan 2025 00:00:00 +0000

From January 19 to 22, two members of i3lab, Pery and Emilio, participated in the Winter School on Artificial Intelligence and Software Engineering organized by the University of Doha for Science and Technology (UDST) in Doha, Qatar.

This event brought together experts, researchers, and professionals from around the world to discuss the latest advancements in Artificial Intelligence (AI) and Software Engineering (SE), two fundamental disciplines driving emerging technologies. Over four days of lectures, workshops, and hands-on sessions, participants explored key topics such as:

✅ Deep learning and generative models, with applications in computer vision, natural language processing, and content generation.

✅ AI-driven software development, focusing on new methodologies and tools to optimize the creation and maintenance of intelligent systems.

✅ Ethics and regulations in AI, a crucial topic in the responsible implementation of AI models in society.

✅ Integration of AI into critical software systems, including applications in healthcare, security, and industrial automation.

Beyond the academic and technical training, the Winter School provided i3lab with the opportunity to strengthen international collaborations, share experiences with fellow researchers, and explore new ideas for applying Artificial Intelligence in intelligent software development.

The visit to Doha also allowed participants to learn more about the local culture and Qatar’s rapidly growing innovation ecosystem, which is strongly investing in artificial intelligence and digital transformation.

We extend our gratitude to UDST for organizing such an outstanding event and to all speakers and attendees for the insightful discussions. Without a doubt, this experience will bring new perspectives and opportunities for research and development at i3lab.

💡 We keep pushing the boundaries at the intersection of AI and software engineering! 🚀🔬

If you’d like to learn more about our research lines in applying AI to software development, feel free to get in touch with us and we’ll share more details.

Second Working Seminar of the AI4Software Research Network

Thu, 16 Jan 2025 00:00:00 +0000

On January 16th and 17th, we were fortunate to attend the Second Working Seminar of the AI4Software Research Network, organized by our colleagues at Universidad Rey Juan Carlos (URJC), Gregorio Robles and Jesús Gonzalez Barahona.

The seminar was primarily structured around two highly engaging roundtables and a “lightning talk” session to showcase the latest work from each node, ranging from evolutionary algorithms underpinning AutoML proposals to LLM-based agents for requirements engineering. The network’s maturity was evident in the volume and quality of the presented work.

In our case, Chema delivered the “lightning talk” for the University of Extremadura node, presenting, first, our systematic evaluation framework for multi-agent systems based on LLMs for software development and, second, our proposal to organize an AI4Software Summer School in Cáceres, with more details to follow soon.

Additionally, Roberto had the privilege of participating in the roundtable “Where is AI4SE research heading? Expert reflections on trends, emerging areas, and outdated approaches in AI advancements for Software Engineering”. This discussion covered highly relevant topics, such as the application of LLMs in various software development tasks and the need for the network to explore ways to incorporate LLMs (and agents) into undergraduate Software Engineering programs.

To conclude, a demo session was held where, among other projects, the URJC demo on code visualization in augmented reality caught our attention, using, for instance, the city metaphor as in CodeCity. Below, you can see a small web app that allows you to visualize a researcher’s publications with this metaphor using her ORCID. The example uses Tim Berners-Lee’s ORCID.

These were intense but highly rewarding days of work. We return re-energized to continue our research efforts in this field.

If you found this interesting, we encourage you to visit the AI4Software Research Network website.

If you’d like to learn more about our research lines in applying AI to software development, feel free to get in touch with us and we’ll share more details.

A multi-criteria decision support system to evaluate the effectiveness of training courses on citizens' employability

Wed, 01 Jan 2025 00:00:00 +0000

Advancing precision in medical image segmentation: A performance analysis of loss functions for COVID-19 lung infection segmentation in computed tomography images

Fri, 01 Nov 2024 00:00:00 +0000

Precision in Medical Segmentation with Compound Loss Functions

Wed, 25 Sep 2024 00:00:00 +0000

We’re excited to share our latest publication! Our new study explores how different loss functions affect the precision of medical image segmentation using deep learning. Specifically, we evaluate three types—AUFL, DSCL, and CE—applied to the detection of COVID-19-related lung lesions in CT scans, using advanced architectures like TransUNet, ViT, and U-Net.

The results are striking: the compound loss function AUFL stands out for its accuracy and stability, consistently outperforming the others—especially in segmenting small lesions like ground-glass opacities (GGOs), which are common in this condition.

This work marks a milestone in our research line focused on improving deep learning models for clinical-grade segmentation, with real-world applications in AI-assisted diagnosis and treatment. Our goal: to ensure reliable and reproducible segmentations, even in lower-resolution modalities like CBCT.

Advancing precision in medical image segmentation: A performance analysis of loss functions for COVID-19 lung infection segmentation in computed tomography images is one of the outcomes of musicgenia, a project funded by grant CPP2021-008491 from MICIU/AEI/10.13039/50100011033 and the European Union through NextGenerationEU/PRTR.

Curious about why AUFL performs best, and how these loss functions compare across architectures and modalities? Reach out and we’ll be happy to explain.

Analysis of ChatGPT Performance in Computer Engineering Exams

Mon, 25 Mar 2024 00:00:00 +0000

Medical image segmentation with SAM

Fri, 23 Feb 2024 00:00:00 +0000

We are celebrating! Our work dedicated to the analysis of the medical image segmentation capabilities of SAM, the foundational model of Meta, has just been published. In our tests, this model achieves results close to those obtained by models specifically built for this task, even though SAM has not been trained to perform segmentation of this particular type of images.

This is our first publication in the field of MIA. This opens a new line of research within the laboratory, with which we hope to begin a fruitful collaboration with different entities in the field of health.

Delving into the results obtained… would you like to know why they are so interesting? Contact us and we will tell you.

No More Training: SAM's Zero-Shot Transfer Capabilities for Cost-Efficient Medical Image Segmentation

Mon, 15 Jan 2024 00:00:00 +0000

Intel supports our research

Mon, 31 Jul 2023 00:00:00 +0000

We already have a brand-new Intel GPU in our hands! We requested it a few months ago to use it in the different research projects related to Artificial Intelligence that we are working on. We will tell you more about it when we return from the vacation.

Thanks to Intel for trusting our lab and Jocelyn for facilitating the process. Surely this is only the first of many collaborations.

Having Intel is an honor, but being able to thank John Carmack for sharing the details of the program through Twitter is an unexpected extra ball:

Intel is giving away free GPUs for academic work. https://t.co/w5sdEIlkx8
— John Carmack (@ID_AA_Carmack) December 14, 2022

Interested? Want to know more? Contact us, and we’ll talk about it.

JENUI 2023: best paper

Thu, 06 Jul 2023 00:00:00 +0000

On November 30, 2022, OpenAI introduced ChatGPT. Like most of us, we were surprised 🤯 with its incredible capabilities. From the first moment, we wondered what impact it would have on the different facets of our day-to-day life: in Computer Engineering, research, and teaching. Based on an idea from our colleague Roberto, we decided to focus on the latter.

The result of several months of effort is our work “Impact of ChatGPT in the evaluation methods of a Computer Engineering degree “, which we presented to the JENUI of 2023 and for which we have been awarded the prize for best paper of the conference. In this edition, dominated by ChatGPT (something that has kept some speakers awake at night), the paper “To rival or not: analysis of Wooclap’s competition mode based on performance and audio processing “, by our colleagues from the University of Murcia, was also awarded.

Here you can see Roberto presenting our experience to the rest of the attendees (about 10 minutes plus question time):

}

We will share the proceedings as soon as they are published so you can read the work and give us your feedback. In the meantime, here is the abstract:

Generative Artificial Intelligence has undergone an essential evolution during the last years and especially in 2022. One of the intelligence that has caused the most buzz in academia is ChatGPT, which provides an interface that significantly simplifies the productive use of a large language model. These language models can parse and generate text with great speed and quality. These capabilities could have a relevant impact on teaching-learning methodologies and assessment methods. In order to analyze the possible impact of ChatGPT on assessment methods, in this work, ChatGPT’s ability to solve the exams of 15 Software Engineering subjects of a Computer Engineering degree has been tested. The results show that ChatGPT could noticeably impact the assessment methods; it can overcome many questions and problems of different natures in multiple subjects. As a fundamental contribution, a detailed study of the results by type of questions and problems is provided, which allows us to establish recommendations to be taken into account in the design of assessment methods.

Are you interested in the subject? Do you want to collaborate with us? Contact us, and we’ll talk.

Update: You can read the cited work right here.

Impacto de ChatGPT en los métodos de evaluación de un grado de Ingeniería Informática

Wed, 05 Jul 2023 00:00:00 +0000

ADAPIMMA

Tue, 04 Jul 2023 11:31:59 +0200

Copyright management in Spain is an issue that has always been in the eye of the hurricane and that has aroused great controversy. Special mention deserves the management of copyrights of musical pieces. The SGAE (General Society of Authors and Publishers) monopolizes managing these rights. Copyright management is governed by the Intellectual Property Law, which in its wording from 1996, is somewhat vague regarding distribution criteria. The wording could be more precise. It needs to be clarified how the SGAE rates the issuing entities, such as the radio. In some cases, it is done by the billing volume of the issuing entity, applying a percentage. In other words, it is independent of the music used by the station, and, this being the case, there is no way to make a fair distribution among the authors who contribute that music. In other cases, estimates are made for the use of the music, and based on that, the management company bills. This can lead to situations where talented young authors with short and unconsolidated careers need to receive adequate rewards for their work and discourage them from continuing their careers in this field.

Precisely the objective of this project is to investigate the techniques that would allow the development of a system capable of i) discriminating the use made of music in radio media according to its cataloging in the new Intellectual Property Law (solution to the first problem) and ii ) associate the exact use made of the piece of music to its original author in order to make an equitable distribution of royalties (solution to the second problem). The development of this proposal would mean a significant innovation in the musical world that could be an essential source, not only of potential clients worldwide but also of international investments that allow the internationalization of the ICT model of Extremadura, both fundamental lines of the Employment axis, Entrepreneurship and Innovation of the Regional Green and Circular Economy Strategy, Extremadura 2030. Not surprisingly, due to its evolution and potential path in Extremadura, the ICT sector is an active pole of production of transversal technology and a motor for the establishment of software development companies, which can be considered green companies as they do not generate waste.

Results

The paper “A Hybrid Multidimensional Recommender System for Radio Programs”, published in the Elsevier’s “Expert Systems with Applications”, is a direct result of this project.

It’s source code is available in GitHub via two projects:

A demo is also available.

SAM in Medical Image Segmentation

Tue, 23 May 2023 15:00:53 +0200

The project’s source code can be found in this repository.

MusicGenia

Fri, 12 May 2023 13:17:36 +0200

MusicGenia’s main goal consists in building a cloud-based platform to offer AI-generated production music as a service for content creators and media, which can be consumed online (live music generation) or offline (recorded music generation). The direct benefits of that platform are: (1) music free of royalties; (2) original music; (3) easy to find music suiting your content; and (4) music on stream (consume as much as you need, pay per seconds not per song).

Guest Lecture by Graduate Student Mario Corchero

Tue, 29 Nov 2022 17:30:00 +0000

Roberto Rodríguez Echeverría and José María Conejero Manzano have organized an invited talk by UEx graduate student Mario Corchero for current students. Mario is currently working at Bloomberg, and in his talk, he wants to show our students how the subjects he took in the degree have been helpful to solve real problems and challenges he has encountered in his work at Bloomberg. Javier Zugasti will join him to tell you about Bloomberg’s job opportunities.

The talk will take place on Tuesday, November 29, at 17:30h in the Assembly Hall of the Polytechnic School.

Although initially geared towards 3rd and 4th-year students, it is suitable for any student, regardless of year.

Mario Corchero is a Senior Software Developer at Bloomberg. He leads the Python infrastructure team, enabling the company to work effectively in Python, building company-wide libraries and tools. His professional experience resides mainly in Python and C++, and he has contributed some patches to multiple Python open source projects.

Javier Zugasti holds a degree in Telecommunications Engineering from the Polytechnic University of Madrid. After two years working in Germany in the automotive industry, Javier has worked at Bloomberg Engineering for 12 years as a software developer and managing engineering teams producing financial applications for the Bloomberg Terminal. In the last 3 of those years, managing the News Production Engineering department that supports almost 4000 journalists worldwide responsible for creating Bloomberg News content.

Update: the event was a huge success!

If you are interested in their scholarships or jobs, you can find more information here:

Homeria at the first Caceres Talk 2022

Thu, 24 Nov 2022 18:30:00 +0000

On November 24, at 18:30, within the first event Cáceres Talk, Fernando Sánchez Figueroa and Juan Carlos Preciado (co-founders of Homeria) tell the success story of WinDelivery, created by Homeria and now part of Glovo. This talk is highly recommended if you are interested in learning more about transferring knowledge from the university to the company. You’ll know how Extremadura’s spin-offs can compete with multinational companies. In addition, Fernando and Juan Carlos will tell all the steps they have taken along the way, what they learned from them and how this learning has led them to where they are now.

If you would like to attend, [reserve your place][reserve] before the event is full.

Update: the event in Digital Region.

Metrikamedia at Conecta Empresas 2022 Forum

Thu, 24 Nov 2022 09:30:00 +0000

On Thursday, November 24, 2022, within the technological showroom Foro Conecta Empresas, Metrikamedia, a spin-off of the UEx, presents one of the technologies developed in its company. It is used to separate voice and music within radio broadcasts to facilitate their auditing by advertisers and monitoring companies. Visitors can try to identify a piece of music being played in the background while an announcer is speaking, demonstrating the usefulness of this platform.

Attendance at the event is free of charge, and no reservation is required. Take advantage of the opportunity to chat with those responsible for applying Artificial Intelligence techniques in real projects.

National project granted

Thu, 13 Oct 2022 00:00:00 +0000

We’ve been granted a national project of the Knowledge Generation Call 2021 for the development of Reproducible and Replicable Data Science Processes.

Last October we received the news that the Ministry of Science and Innovation had approved our project proposal entitled “Model-Driven Engineering for Data Science Projects “, led by José M. Conejero. Thanks to this, we will be able to investigate new techniques to improve the reproducibility and replicability of data science projects through the application of concepts related to Model-Driven Software Development. Also, through the application of these techniques, it will be possible to face some of the current challenges of this type of projects, such as fairness or data provenance. With all this, this project will help to alleviate the problem known as the reproducibility crisis, which started in the branches of medicine and psychology.

You can find more information about the awarding of the projects of this call in the final resolution proposal of the procedure for awarding grants 2021 to “PROYECTOS DE GENERACIÓN DE CONOCIMIENTO” in the framework of the Programa Estatal para Impulsar la Investigación Científico-Técnica y su Transferencia, del Plan Estatal de Investigación Científica, Técnica y de Innovación 2021-2023.

Deep Learning Music Generation

Mon, 03 Oct 2022 00:00:00 +0000

On November 3 and 4, as part of the Ph.D. Program in Computer Science at the University of Almeria, Roberto Rodríguez Echeverría, University of Extremadura, taught the course Generation of content and/or recognizable patterns using neural networks, with the participation of Jorge Periánez Pascual, MetrikaMedia.

For some time now, relying on artificial intelligence to help us in our daily work has ceased to be a dream and has become a reality. Every day we use machine learning models to suggest the next words we will type on our cell phones, to give just one example. What we would never have imagined is that this artificial intelligence would help us to create creative content such as code for our programs, images following textual descriptions, or, as we explained in this seminar, music", commented UEx professor Roberto Rodríguez Echeverría.

The content of the seminar was as follows:

Introduction to Deep Learning
Recurrent Networks and Transformers
Content Generation
Image Generation with Variational Autoencoders
Music Generation with LSTM and Transformers

This and other seminars are the results of the collaboration between the company and the University. From i3lab, one of the working groups of Quercus Software Engineering Group, we are committed to the training of new engineers in areas such as artificial intelligence and everything related to it, such as MLOps. Interested? Contact us and let’s talk.

A Model-Driven Approach for Systematic Reproducibility and Replicability of Data Science Projects

Tue, 28 Jun 2022 00:00:00 +0000

CompareML: A Novel Approach to Supporting Preliminary Data Analysis Decision Making

Wed, 01 Jun 2022 00:00:00 +0000

Suggesting model transformation repairs for rule-based languages using a contract-based testing approach

Tue, 01 Feb 2022 00:00:00 +0000

A model-driven approach for the definition of reproducible and replicable data analysis projects

Sat, 01 Jan 2022 00:00:00 +0000

SimulateIoT: Domain Specific Language to Design, Code Generation and Execute IoT Simulation Environments

Fri, 25 Jun 2021 00:00:00 +0000

Applying data driven decision making to rank vocational and educational training programs with TOPSIS

Mon, 01 Mar 2021 00:00:00 +0000

A Hybrid Multidimensional Recommender System for Radio Programs

Sat, 20 Feb 2021 00:00:00 +0000

A Real-Life Machine Learning Experience for Predicting University Dropout at Different Stages Using Academic Data