Authors - John Jenq Abstract - An Artificial Intelligence (AI) agent is an expert that can handle tasks efficiently and accurately, and can execute specific tasks automatically. In this paper, we develop an AI class scheduling assistant system. This system consists of a large language model (LLM) which serves as an interface to the user, and an agent. This agent contains several sub-agents to perform the scheduling task. The data used by the agent was acquired from the University website. It was pre-processed and stored on an excel file that is accessible to the agent. One of the sub-agents will load the file and convert it into Python Pandas data frame. Another sub-agent has the function of finding all the available class sections based on the user’s input from the data frame. This available section information can be used later, along with the user’s data by other sub-agents. OpenAI GPT-4 model is used for the LLM. We implemented the system using Python programming language. According to our experimental results, the agent workflow runs smoothly and quickly, and the system performs the scheduling task accurately.
Authors - Anh Truong, An Nguyen, Huy Le, Thang Pham Abstract - Large language models (LLMs) are being increasingly deployed in customer-service chatbot applications. While traditional fine-tuning approaches often suffer from hallucinations or require computationally expensive retraining, emerging retrieval-based and parameter-efficient methods are increasingly regarded as promising alternatives. However, comprehensive evaluation of these paradigms in a customer-service context remains limited. To address this gap, we conduct a comprehensive comparison of three fine-tuning paradigms - Retrieval-Augmented Generation (RAG), Retrieval-Augmented Fine-Tuning (RAFT), and Weight-Decomposed Low-Rank Adaptation (DoRA) - for customer-service chatbot applications. All methods share a common backbone and are evaluated on the Bitext customer-support dataset. RAG method achieves strong factual consistency at the cost of higher inference latency. RAFT delivers the best overall balance of intent-classification accuracy, coverage, and low hallucination with moderate latency overhead. DoRA extends LoRA (Low Rank Adaptation) by decomposing weight updates into magnitude and directional components for precise low-rank adaptation. On Bitext, DoRA’s end-to-end fine-tuning not only achieves medium accuracy and coverage but also suffers a higher hallucination rate due to limited training data. In larger-scale settings (e.g., SQuAD + 400K-row FAQ), DoRA demonstrates surprising low hallucination and high BLEU/ROUGE, proving that dataset size significantly influences its performance. The results highlight trade-offs between latency, accuracy, and factual reliability, while emphasizing the importance of data scale and retrieval grounding in deploying scalable, trustworthy LLM-based customer-service systems.
Authors - Tahfizul Hasan Zihan, Hana Sultan Chowdhury, Shirazim Munir Deap, Rubayed Mehedi, Farhad Alam, A. M. Shahabuddin, Mahady Hasan Abstract - Recirculating Aquaculture Systems (RAS) have increasingly gained attention in Bangladesh due to their efficient water use. However, maintaining real-time water quality is challenging, as delayed anomaly detection often causes economic loss and compromises fish health. To address this, we developed a low-cost IoT-based monitoring and alert system that integrates industry-standard sensors to track key water parameters (pH, turbidity, dissolved oxygen, temperature, oxidation-reduction potential, total dissolved solids, and electrical conductivity) using Arduino microcontrollers and ESP8266 Wi-Fi modules. The total hardware cost was under USD 700, an order of magnitude cheaper than commercial systems. Sensor readings were published every 15 seconds via MQTT to the Thing Speak cloud platform, enabling real-time visualization and WhatsApp-based alerts for immediate intervention. Experimental deployment demonstrated the system’s capability to maintain optimal conditions consistently, identifying critical events such as abnormal water quality drops efficiently. Clear daily and monthly trends were observed, enabling predictive adjustments and informed automated decision-making. This IoT approach significantly improves operational efficiency and reliability in resource-limited aquaculture settings typical of Bangladesh, ensuring sustained fish productivity and welfare.
Authors - Ayman Alarabiat, Yousef Alarabiat, Mahmoud AlZuabi, Mamoun Shakatreh Abstract - Governments are increasingly leveraging artificial intelligence (AI) Chatbots to enhance e-service accessibility. However, Chatbots adoption among citizens’ remains low, limiting their intended benefits. This study explores citizens’ perspectives on government Chatbots adoption. A quantitative correlational research approach was employed, collecting 358 responses from Jordani-an citizens who had used government Chatbots in the last 6 to 12 months. An online survey measured ten key constructs: low complexity, relative advantage, compatibility, trialability, observability, trust, responsiveness, perceived intelligence, anthropomorphism, and Chatbot adoption. Data analysis using SPSS 24 revealed low willingness to adopt Chatbots, primarily due to concerns about Chatbot intelligence, responsiveness, and trust. Additionally, low observability and limited perceived relative advantage further hinder adoption. These findings provide insights for policymakers, government agencies, and Chatbot developers to enhance Chatbot functionality and user experience. Key recommendations include improving Chatbot intelligence and responsiveness, increasing public awareness, and fostering greater trust in the Chatbot. Addressing these factors can drive greater adoption, maximizing the efficiency and impact of AI-driven public services.
Authors - Maisyaroh, Agus Timan, Mustiningsih, Maulana Amirul Adha, Indra Lesmana, Rudy Ansar, Novia Putri Arianti Abstract - Digital learning transformation in rural schools faces various challenges, including limited infrastructure and low digital literacy among teachers. This study aims to explore the contribution of collegial supervision in supporting this transformation. Using a descriptive qualitative approach, data were collected through in-depth interviews, observations, and documentation at a high-performing private secondary school in a rural area. The findings reveal that collegial supervision is grounded in values of trust, equality, shared reflection, and communal cooperation. This approach provides a collaborative space for teachers to exchange best practices, enhance their technological competencies, and develop effective digital teaching strategies. Despite challenges related to time, resources, and varying levels of understanding, collegial supervision has proven effective in fostering an innovative school culture that supports digital-era teaching and learning. This study contributes to the literature on educational supervision by offering context-specific insights and practical strategies for teacher development in resource-constrained schools.
Authors - Nishat Shaikh, Parth Shah, Bimal Patel Abstract - Deep learning has revolutionized oncology by enabling unprecedent-ed integration of multimodal data for cancer diagnosis and prognosis. This comprehensive survey presents the first systematic analysis of deep learning architectures across four major cancer types (lung, breast, skin, and brain) through three critical data modalities: medical imaging, histopathology, and genomics. Our unique contribution lies in providing a structured taxonomy of multimodal fusion strategies and identifying critical architectural innovations that have emerged in the 2021-2025 period. We systematically analyze 60+ recent studies, revealing that attention-based mechanisms and Transformer architectures demonstrate superior performance in handling heterogeneous cancer data compared to traditional CNN approach-es. Our analysis uncovers three key research gaps: (1) limited interpretability frameworks for clinical deployment, (2) insufficient standardization across institutions, and (3) scalability challenges for real-world implementation. This survey uniquely bridges the gap between theoretical deep learning advances and practical oncological applications by proposing a unified framework for multimodal cancer analysis. We provide actionable insights for researchers and clinicians, establishing clear directions for future development in AI-driven cancer care that addresses both technical innovation and clinical translation requirements.
Authors - Esthefano Palomino, Rodrigo Condor, Edgar Ramos, Ludwig Tocto, Victor Pimentel Abstract - Agri-food supply chains are increasingly exposed to climate variability, resource constraints, and disruptive shocks, requiring strategies that enhance resilience and sustainability. This study applies the Best–Worst Method (BWM) to assess and prioritize four strategic dimensions: resilience, smart manufacturing, innovation, and circularity. The results indicate that resilience is the most critical enabler, while smart manufacturing and innovation provide significant complementary support, with circularity ranking lowest. These findings emphasize the central role of resilience in safeguarding agri-food systems and demonstrate how technology and innovation can strengthen long-term sustainability. The study contributes by offering a structured decision-making framework that helps managers and policymakers focus on the most impactful dimensions when designing more adaptive and sustainable agri-food supply chains.
Authors - Sarika Pabalkar(Wagh), Kirti Jain Abstract - Generative adversarial networks, or GANs, are strong tools for improving medical pictures because they can make high-quality virtual images that can be used to solve problems like limited datasets, image unpredictability, and poor diagnosis accuracy. This research looks at all the ways that GAN can be used in medical imaging and highlights the most important improvements in the process of segmentation reconstruction, disease identification, and cross-modal synthesis. A structured methodology based on PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) was employed to identify, screen, and analyze 42 peer-reviewed articles published between 2020 and 2025 across databases such as PubMed, Scopus, IEEE, and ScienceDirect. The review compares widely used GAN architectures—including CycleGAN, Pix2Pix, DCGAN, and ProGAN— evaluating their strengths, limitations, and suitability for different imaging modalities such as MRI, CT, X-ray, ultrasound, and mammography. There are significant problems, including mode collapse, instability in training, no consistent evaluation standard, and few opportunities for clinical validation. The results show the increasing importance of GANs in generating clinically useful data for rare disease information and for cross-modality tasks. The discussion also indicated future interventions that would focus on model stability, ethical use, clinical incorporation, and generalization across populations. The aim of the review is to help fellow researchers and practitioners through assessing state-of-the-art GAN approaches and defining gaps that will need to be addressed prior to more adaptation of GANs in medical imaging.
Authors - Om Roy, Dhruv Shingala, Priyanka Patel Abstract - Bias within Artificial Intelligence (AI) systems constitutes a profound challenge with substantial implications for fairness, accountability, and societal equity. This paper presents an exhaustive examination of the ontology of bias in AI, delving deeply into its conceptual underpinnings and exploring the intricate algorithmic consequences arising from biased data and models. By establishing a comprehensive and nuanced framework that categorizes diverse manifestations of bias and elucidates their origins, this study aims to foster a profound understanding of how bias permeates AI systems. Integrating interdisciplinary perspectives drawn from philosophy, sociology, and computer science, the ontology of bias is meticulously dissected to reveal its multifaceted nature. Furthermore, the paper investigates the profound impacts of these biases on critical decision-making processes and proposes multifaceted strategies for mitigating bias through ethical design, advanced algorithmic techniques, and stringent regulatory frameworks. Through detailed case studies and empirical analysis, this research highlights the inherent complexities in addressing bias and underscores the imperative for collaborative endeavors to cultivate equitable AI technologies.
Authors - Payel Das, Tejaswini Seelam, Rajeswari Annam Abstract - This study investigates the intersection between generative AI, sustainable development and education by conducting a systematic review on the utilization of ChatGPT and other Large Language Models (LLMs) in relation to Sustainable Development Goals (SDGs), focusing on SDG 4 and SDG 12. The article explores how technology-enhanced learning and LLMs are challenging pedagogical traditions, practices, even ethical dilemmas, and calls for responsible governance. There were 59 peer-reviewed articles included between 2023-2025 which provided data synthesis. Three core clusters were discerned using the VOS viewer keyword co-occurrence map: (1) educational impact and learning outcomes, (2) stakeholder ethics and governance perspectives, and (3) integrity Challenges of AI-assisted instruction. The study also raises alarms about threats that could emerge from misinformation, academic cheating, algorithmic bias and unequal ac-cess. It also underscores the lesser-known environmental footprint of AI tools.
