UPLB ICS Peak One

This paper presents a new approach in implement- ing lip reading systems by using real time facial landmark detection models. While there are significant advancements in the development of lip reading systems for high-resource languages such as Chinese and English, the progress for the development of a robust Filipino lip reading system has been slow due to the lack of comprehensive datasets. To address the gap, our approach is to create a model that is tailored to a relatively low amount of data. The researchers used Google’s MediaPipe Face Mesh to track 80 landmarks on the lips, capturing detailed movement over time. The data goes through temporal and spatial normalization then is passed through a lightweight 3D-ResNet-TCN architecture, de- signed to extract spatio-temporal features while being computa- tionally simple. The model achieved test accuracies up to 76.60% across 30 Filipino words, a significant improvement over previous video-only Filipino lip reading models. Words with distinct bilabial sounds and clear vowel transitions were recognized with higher accuracy, while those with similar lip movements or non- labial sounds presented challenges. This research demonstrates that landmark-based approaches effectively support lip-reading in low-resource contexts, improving communication accessibility for people with hearing impairments.

This study presents the development of a web- based knowledge management system that integrates artificial intelligence techniques to enhance knowledge management within software development teams. The primary aim was to implement Retrieval-Augmented Generation, enabling effective storage, re- trieval, and automatic generation of explicit knowledge. The system incorporates semantic search capabilities and automated workflows for validating knowledge entries, ensuring standard- ization and preventing redundancy. Version control is managed through integration with GitHub, facilitating precise tracking and maintenance of the knowledge base. Evaluation using the System Usability Scale resulted in an average usability score of 86.84, indicating excellent usability and significant potential for improved collaboration, documentation quality, and development efficiency within software organizations.

This project addresses the educational challenges faced by K to 4 learners in the Philippines, particularly in science and mathematics, by leveraging the increasing popularity of gaming and internet access. The primary objective was to develop and evaluate story-based and game-driven learning applications designed to enhance student engagement and foundational skills. Two Android applications were created using Unity, C#, and Ink: “Caloy’s Karinderya” for Grade 4 science (balanced diet) and “Fruit Farmer” for Kindergarten mathematics (sums of ten). Both applications integrate narrative elements with interactive games incorporating real-world scenarios. Usability was assessed using a modified System Usability Scale (SUS), and student engagement and motivation were measured with an Enjoyment Scale, each tested with 10 target-age respondents. Results in- dicated high usability and engagement for both applications. While the applications were found to be easy to use, enjoyable, and motivating, some learners still expressed a preference for traditional books. The study concludes that these applications are effective tools for learning, demonstrating the potential of technology to create engaging educational experiences.

People learn about service interruptions through various means, such as news reports, social media, or when an interruption occurs at the moment. The absence of essential services can disrupt daily lives, but with prior notice, effects can be minimized. Given that information is dispersed throughout online channels, a web application, Nona, was developed to centralize and organize interruption announcements. Nona was created to show details of water, electricity, and internet service interruptions on a map. It is powered by a fine-tuned DistilBERT model with an accuracy of 95.66%, extracting information from posts retrieved near real-time from selected public pages on social media. The application includes filtering options for data visualization based on temporal, location, provider, and date details. It also features a notification system to keep users informed. Achieving a System Usability Scale assessment score of 90.81, the application demonstrates excellent usability.

In arts and design, a significant collection of outputs, such as drawings, is considered a product of artists to be promoted—a portfolio. However, creating and managing one can be challenging, especially when most existing portfolios are web-based and have monotonous and repetitive formats; thus, standing out is nearly impossible in a crowded online space. VanishingPoint is a web application that provides users with a hyper-interactive portfolio through the incorporation of a virtual environment in order to increase user immersion and diverge from common portfolio layouts. There are 25 participants for the evaluation to test its usability. The participants selected in this activity are artists, teachers, and students, as they are the potential target users in this project to craft their immersive portfolios and exhibit their artworks through virtual galleries. The evaluation of the system was conducted through the SUPR-Q, which revealed promising results: usability (4.68), trust (4.6), appearance (4.72), and loyalty (4.4). The overall SUPR-Q score is 4.6, which classifies the application to be at the top 1% websites in terms of user experience. VanishingPoint is viewed as a forward- thinking solution and is likely to evolve into a new creative standard, despite some preference for conventional portfolios in professional settings.