en es

Home

Facundo Roffet


🔱 About me

Hi, I'm Facu, an electronic engineer who realized that circuits weren't his thing. In a moment of complete serendipity during the pandemic, I came across the YouTube channel DotCSV and almost instantly discovered that my true passion was artificial intelligence (particularly deep learning). So I decided to start learning about it on my own (I can’t recommend the FastAI course enough).

Fortunately for my career, the National University of the South (Universidad Nacional del Sur - UNS) has the Image Science Laboratory (Laboratorio de Ciencias de las Imágenes - LCI), a nationally renowned institution for research, development, and technology transfer in image processing and remote sensing. There I met Claudio Delrieux (LCI director), who quickly became my mentor and introduced me to Gustavo Patow. The three of us outlined a work plan for me to apply to the CONICET Internal Doctoral Scholarship for Strategic Topics. After obtaining it in 2023, I have specialized in using deep learning and fMRI data to better understand the human brain, particularly in stroke cases.

Beyond my work, I have random interests in other disciplines such as philosophy, geography, and psychology. I’m a strong advocate of learning for its own sake, not just as a means to an end, so I spend part of my free time watching online courses and taking notes. My notes are primarily designed for my own understanding, but I make them available in the Course Notes section in case they’re helpful to anyone else.

I'm also a big fan of nature and adventure. Whenever I get the chance, I travel to discover new places—be it solo, with friends, family, or folks from the Club Andino Bahía Blanca. Additionally, at @stratocurious I share photos of clouds while I learn to classify them.

📖 Learn | 🌍 Explore | 🫂 Connect


🏢 Industry projects

Viterra (January 2025 – April 2025): Conceptualization and development of an application that allows operators to automate the quality control process in seed hulling machines, improving measurement consistency and freeing up valuable operational staff time. The application contains a regression model that, based on photographs of seed samples, predicts the percentages of whole seeds (MAE: 1.50) and damaged seeds (MAE: 3.09).

SpaceSur (November 2023 – December 2024): Development of pipelines to process satellite images. Trained models on public datasets for building segmentation (mAP@0.5: 0.8140) and landfill classification (F1 score: 0.8897). Deliver of a modular codebase to facilitate and streamline future iterations with customer-specific data.

Practia (July 2022 – February 2024): Technical consulting and mentoring. Strategic guidance for a novice team to address and execute their first deep learning projects, shortening their learning curve and ensuring the adoption of best practices.

ECODOC (May 2022 – April 2024): Preprocessing and organization of a dataset of more than 3,000 panoramic dental X-ray images. Model training for tooth segmentation (mAP@0.5: 0.9230) and detection of pathologies such as caries (F1 score: 0.8504) and retentions (F1 score: 0.9127), with the goal of subsequently integrating them into an on-demand software service.


🔍 Academic projects and publications

Doctoral project (April 2023 – Present): Study and development of deep learning–based tools for simulating and predicting whole-brain neuronal activity and its evolution in disease states by defining biomarkers. The goal is for these tools to predict treatment effects and drug responses in neurodegenerative diseases (Parkinson’s, Alzheimer’s, etc.) to contribute to personalized diagnosis.

  • Deep Learning and Whole-Brain Networks for Biomarker Discovery: Modeling the Dynamics of Brain Fluctuations in Resting-State and Cognitive Tasks. Scientific Reports, 2025. LINK.

MultiCaRe project (January 2024 – July 2025): Use of hierarchical ontologies and semi-supervised learning techniques to improve both label quality and data structuring in a medical image dataset of over 150,000 samples across 140 hierarchical categories.

  • An Open-Source Clinical Case Dataset for Medical Image Classification and Multimodal AI Applications. Data, 2025. LINK.
  • The Multiplex Classification Framework: Optimizing Multi-Label Classifiers through Problem Transformation, Ontology Engineering, and Model Ensembling. Applied Ontology, 2025. LINK.

Erythrocytes project (October 2022 – June 2024): Design of a methodology to automatically detect five anomalies in avian red blood cells using an imbalanced dataset to enable rapid, less subjective analysis (F1 score: 0.7703).

  • Integrating Deep Learning into Genotoxicity Biomarker Detection for Avian Erythrocytes: A Case Study in a Hemispheric Seabird. Mathematical and Computational Applications, 2024. LINK.

Undergraduate thesis (September 2021 – August 2022): Use of information theory to evaluate fMRI harmonization techniques, finding that common methods do not completely eliminate site-specific biases.

  • Assessing Multi-Site rs-fMRI-Based Connectomic Harmonization Using Information Theory. Brain Sciences, 2022. LINK.


💻 Software packages

DLOlympus: A Python library with tools and templates to streamline deep learning model training and inference. Adds extra functionality to FastAI and MMDetection. LINK


🎓 Education

PhD in Engineering (April 2023 – Present):

  • National University of the South, Bahía Blanca, Argentina

Electronic Engineering (March 2017 – December 2022):

  • National University of the South, Bahía Blanca, Argentina


📜 Additional training

International Summer School on Artificial Intelligence (2026)

  • National University of Hurlingham, Villa Tesei, Argentina.
  • National University of San Martín, Villa Chacabuco, Argentina.


🎤 Conference presentations

VII Congress of Science and Technology (2024):

  • Introducing the Erythrocytes project and leading the workshop Image classification with AI: fundamentals, best practices, and a hands-on example. National University of Itapúa, Encarnación, Paraguay.

30th AUGM Young Researchers Conference (2023):

  • Presentation of the undergraduate thesis. National University of Asunción, Asunción, Paraguay.

VI IEEE Biennial Congress of Argentina – ARGENCON (2022):

  • Presentation of the undergraduate thesis. National University of San Juan, San Juan, Argentina.


📩 Contact

Affiliated with the Image Science Laboratory at the National University of the South and CONICET.