UPLB ICS Peak One

Capture-the-flag (CTF) challenges are word prob- lems that are presented to the player alongside a vaguely-worded hint. Past attempts to build a CTF solver with LLMs have relied on multiple prompts and human intervention. The current literature suggests it is possible to enhance a large language model’s reasoning capabilities using instruction fine-tuning and chain-of-thought (CoT) prompting. A Llama 3 8B model was fine-tuned to classify four types of hints over a CoT instruction dataset. The fine-tuned model outperformed GPT-4 and the base Llama 3 8B when tasked to classify real-world CTF hints. An agent was built around the model by connecting a GPT-4 critic and a CyberChef ”magic” toolkit.

The challenges surrounding maritime boundary dis- putes have significant implications for the livelihood of fisherfolk communities. The absence of proper navigation assistance exposes them to substantial risks, stemming not only from natural maritime hazards but also from legal conflicts over territorial boundaries. Considering these concerns, this research aims to create a mobile application that utilizes an alert and tracking system. A website application is also developed for administrators to efficiently manage data and promptly respond to distress calls. System Usability Testing (SUS) was conducted with 20 respondents. Ten users for testing each of the applications. The mobile application has 92 out of 100 SUS scores while the website application has 86.5 out of 100 SUS scores. This indicates that the developed applications are usable for the intended users.

Subtitles enhance video accessibility for people who are deaf or hard of hearing by providing a written version of the verbal dialogue. They lack, however, in conveying the emotional nuances of the speaker. A more comprehensive ap- proach that includes speech content and emotional context can promote an engaging viewing experience. This study proposes using Google Cloud Speech Application Programming Interface (API) for speech recognition and the fusion of two models for emotion recognition: a Convolutional Neural Network (CNN) model for Audio Emotion Recognition and transfer learning with the ResNet-50 model for Visual Emotion Recognition. A mobile application is developed as the interface for the users. The speech recognition API achieved a word error rate of 13.78%. For emotion recognition, the audio emotion recognition model achieved a top-1 accuracy of 65.17% and a top-3 accuracy of 88%, while the visual emotion recognition model achieved a top- 1 accuracy of 64.29% and a top-3 accuracy of 90.10%. The fusion of both models resulted in a top-1 accuracy of 76.33% and a top- 3 accuracy of 95.70%. The application got a mean SUS score of 81.60. Overall, the findings of the study highlight the potential of combining speech and emotion recognition with the use of CNNs.

SaveSmart is a mobile-based financial tracker ap- plication that can be used on both Android and iOS devices. It has essential functions in managing personal finances like adding wallets, recording transactions, showing transaction stats, setting financial goals, recording debts and money lent, and allocating budgets. The application was tested by 23 users and evaluated using the System Usability Scale (SUS), where it achieved a score of 84.58, indicating excellent user satisfaction.

The manual pen and paper enrollment process of Libon Community College (LICOM) is tedious, resource exten- sive, and less reliable for its stakeholders. This study developed a pre-enrollment system that automates the enrollment workflow of LICOM to improve their enrollment process. The quality of the system is evaluated through a modified version of the ISO 25010 standard by a sample of students, instructors, clearance signa- tory, and registrar. The SUS, security, performance efficiency, and functional suitability scores of the system are 75.25, 4.17, 4.15, and 4.34, respectively. This result indicates that eLICOM has good software and system quality in terms of usability, security, performance efficiency, and functional suitability; thus, the application is effective in improving the enrollment process of LICOM.