Conference Agenda

Precision livestock farming in extensive rural environments faces significant constraints related to connectivity availability, energy resources, and communication latency, which limit the adoption of centralized cloud-based architectures. This work proposes an embedded and autonomous Edge AI architecture for livestock monitoring, designed to operate continuously in scenarios with limited resources. The architecture integrates low-power sensors, embedded processing, and a lightweight artificial intelligence model that performs local inference, enabling autonomous decision-making at the node level and an event-based selective communication strategy. As a case study, the proposed architecture was implemented and evaluated at a prototype level in a representative extensive livestock farming scenario, analyzing its performance in terms of inference latency, energy efficiency, and reduction of transmitted data volume. The obtained results show inference latencies on the order of milliseconds, a reduction greater than 80% in communication volume compared to centralized approaches, and operation compatible with low-power consumption schemes. These findings demonstrate the technical feasibility of embedded Edge AI as a scalable and sustainable solution for livestock monitoring in rural environments with limited infrastructure.

This study presents the participatory design and implementation of an ERP system called Micro-Gestión, developed to automate critical processes in micro-businesses in Cali, Colombia. The research is based on evidence highlighting the high vulnerability of these productive units due to manual administrative practices and low levels of technological adoption. To address this issue, a hybrid methodology was employed, combining agile development with user-centered co-creation principles and integrating the Technology Acceptance Model (TAM) as an evaluation framework. The system was built on the open-source Odoo platform, leveraging its modular architecture to customize three essential components: CRM, Sales, and Invoicing. Requirements were gathered through three participatory workshops involving four micro-entrepreneurs, who actively contributed to defining workflows, simplifying processes, and validating prototypes. The technical development was carried out by Information Technology students using Python, PostgreSQL, Visual Studio Code, and GitHub under an iterative two-week sprint framework. The results demonstrate significant improvements in operational efficiency. Customer registration time was reduced by 86%, sales processing by 56%, and invoice generation by 95% compared to previous manual methods. The five-step integrated workflow optimized the operational sequence and decreased process errors by 68%. Post-implementation evaluation using a Likert scale showed an average acceptance score of 4.46 out of 5, confirming a high perceived usefulness and ease of use.

Agriculture faces critical challenges—climate change, food security, and resource degradation—that demand innovative solutions. This study analyzes, through a systematic literature review (46 articles, 2020–2025), the impact of artificial intelligence (AI) on agricultural systems, evaluating technologies such as machine learning, the Internet of Things (IoT), and precision agriculture. The results demonstrate that AI significantly optimizes productivity, with increases of up to 30% in fertilizer efficiency [1] and reductions of up to 40% in labor costs [2] and environmental sustainability. However, its adoption is still limited by the weight of economic barriers (high initial costs) and social barriers (mistrust, lack of training) [3], [4]. In order to analyze these dynamics, a simulation model based on Forrester's system dynamics is developed, which highlights three critical factors in the adoption of this technology: Public investment in new digital infrastructure; Training programs for rural communities; Ethical strategies that prioritize transparency and community participation [5],[6]. The study concludes that the effective integration of AI into agriculture requires a sustainable and adaptive approach to public policies that reduce technological gaps and promote resilient agricultural systems

This study aims to compare traditional mathematical models used to analyze the spread of infectious diseases with machine learning methods, using COVID-19 as a case study. Classical models such as SIR, SEIR, and SIS help describe the progression of an epidemic; however, they typically rely on fixed parameters and struggle to adapt to real-time changes. In contrast, machine learning models including Polynomial Regression, SVM, and Random Forest are capable of processing large datasets, detecting more complex patterns, and adjusting their predictions as new information becomes available. For this analysis, historical data from the World Health Organization (WHO), adapted to national records, were used, and the performance of each model was evaluated using metrics such as RMSE and R². Overall, the results showed that machine learning models provided a better fit and greater adaptability, making them a valuable option for anticipating and controlling future epidemic outbreaks.

This paper presents the development of a mobile application for automated skin imperfection classification using deep learning techniques. The system integrates Google Colab for model training with Hugging Face's pre-trained models, implementing transfer learning to classify five distinct dermatological conditions: acne, rosacea, eczema, psoriasis, and atopic dermatitis. The trained model is deployed through an Android application developed in Kotlin, enabling real-time diagnosis from smartphone cameras. The implementation leverages cloud-based GPU resources for efficient training while maintaining a lightweight mobile interface for accessibility in resource-limited settings. Performance metrics demonstrate high accuracy in disease classification, with potential applications in early detection and telemedicine scenarios. The system architecture emphasizes scalability, user privacy, and clinical utility, providing a practical tool for preliminary dermatological assessment. Achieving an accuracy of 85%.

In rural areas of Peru, early detection of skin cancer remains a critical challenge due to the limited availability of specialized dermatological services. To address this need, a mobile application powered by artificial intelligence was developed to analyze skin lesion images and estimate the probability of malignancy. The system is based on MobileNetV2 using Transfer Learning, trained with the Skin Cancer: Malignant vs Benign dataset from Kaggle and optimized through conversion to TensorFlow Lite for on-device execution in Android environments. The model achieved a validation accuracy of approximately 82%, demonstrating its potential as an accessible tool to support initial clinical screening and assist users in seeking timely medical evaluation.