International Journal of Artificial Intelligence

RAG-Guardrails Integration for AI Content Control

Rakesh More — 2025-12-23

Generative AI is particularly Large Language Models (LLMs), has shown remarkable potential across domains such as healthcare, legal services, and finance. However, their adoption is hindered by two persistent challenges: hallucination, where models generate factually incorrect information and the risk of producing biased or unsafe content. This paper proposes a hybrid framework that integrates Retrieval-Augmented Generation (RAG) with NVIDIA NeMo Guardrails to address these concerns. RAG mitigates hallucinations by grounding model outputs in externally retrieved, trusted data sources, while NeMo Guardrails enforce domain-specific safety and compliance constraints through predefined behavioral policies. Empirical evaluations demonstrate that this combined approach reduces hallucinated content by 30–45% and improves safety and policy adherence across multiple enterprise use cases. The system exhibits strong potential for deployment in regulated, high-stakes environments. Future work will focus on enhancing real-time responsiveness and expanding multilingual and culturally adaptive capabilities. The proposed framework offers a scalable foundation for building trustworthy, domain-aligned generative AI solutions.

Design and Evaluation of a Fuzzy Logic Based Intrusion Detection System for Network Security

Ayomitope Isijola — 2025-12-23

With the proliferation of networked systems, intrusion detection systems (IDS) have become vital in identifying and mitigating cyber threats and unauthorized access. Traditional IDS approaches, such as signature-based and anomaly-based methods, often struggle to detect novel attacks and tend to generate high false alarm rates. This study presents a robust, fuzzy logic-based IDS designed to detect network intrusions and assess their risk levels while minimizing false positives. The IDS classifies network intrusions by analyzing parameters such as source bytes, destination bytes, and packet rates, categorizing them into risk levels through defined fuzzy rules. Implemented in Python using libraries like scikit-fuzzy and pandas, the system utilizes the KDD Cup 99 dataset, a widely recognized IDS benchmark. Fuzzy membership functions and inference rules were defined for the primary input variables, enabling the system to infer intrusion likelihood. The IDS was tested using both two-variable and multi-variable input setups. It achieved a precision of 0.89, a recall of 0.85, and an F1-score of 0.87 in the multi-variable scenario. Results indicate that the fuzzy logic-based IDS achieves a balanced trade-off between detection accuracy and interpretability. It offers a transparent decision-making framework suitable for real-time applications due to its adaptability and potential for integration with live data streams. This research proposes future improvements by creating a foundation for hybrid intrusion detection systems (IDS) that integrate fuzzy logic and machine learning to enhance accuracy and interpretability. It recommends future research on adaptive fuzzy rules, real-time data processing, and explainable AI (XAI) to improve system flexibility, responsiveness, and transparency in cybersecurity applications.

Automatic Pose Recognition in Basketball Videos Using Entropy, Mean and Standard Deviation

Aliga Paul — 2025-12-23

Most existing models for automatic action recognition in basketball videos lack privacy-friendly analytics, versatility and explainability. So, coaches, players and analysts often invest substantial resources by relying heavily on visual appearance, ball tracking and court context. Unfortunately, this method can be resource-intensive and potentially susceptible to unforeseeable intrusions. This study proposes an entropy-based analytical model for automatic recognition of key basketball actions, designed to optimize the video review process to address the above limitations. The model is implemented with Python programming language to analyze entropy arrays, the mean and standard deviation values derived from 22 basketball game videos. Evaluation suggests that the model flagged basketball_Video2, Video3 and Video9 as containing key moments deserving closer inspection. This has successfully reduced the input datasets to just three critical videos (with mean and standard deviation pairs of 1.96 & 0.33, 2.05 & 0.31, and 1.94 & 0.20) that warrant detailed examination. This targeted filtering significantly improves review efficiency by conserving time and resources and effectively eliminated 19 videos deemed redundant or of lower priority. The approach demonstrates high precision in identifying impactful gameplay moments and addresses a long-standing challenge with workload reduction in basketball analytics without sacrificing review accuracy. Consequently, this method not only supports privacy-conscious analytics but also provides coaches, players and sports analysts with a more focused, resource-efficient framework they can adopt for performance evaluation and strategic decision-making in basketball.

Student Expense Tracking System Using OCR

Ahmad Fadli Saad — 2025-12-23

Nowadays student schedules are packed with their academic and curricular activities. Therefore, Students are no longer tracking their expenses because it is so hard to keep track with their expenses when they a have busy life. The aim of this research is to help students easily track their expenses by automating the process of extracting information from receipts. This research presents a student tracking expenses system using Optical Character Recognition (OCR) technology. The method that was used to develop the system was Website Development Life Cycle (WDLC). The system also uses Image Processing that implements OCR into the system. The system has been tested with a set of sample receipts, and the results show that it is able to accurately extract the relevant information with a high level of efficiency. The initial of this research involved designing the system, which was achieved through the creation of a detailed mockup and wireframe to establish a clear vision for its design. Then, it focused on developing the system, incorporating OCR technology to extract text from receipts. Thorough functional testing ensured that all system features, including user identification, image upload and OCR processing, expenditure management, budget setting, and data visualization, functioned as intended. The system offers users accurate and dependable capabilities for spending pattern analysis, budget management, and expense monitoring. Furthermore, the usability testing was conducted using the Post-Study System Usability Questionnaire (PSSUQ) from 30 students. The mean score of the System Usefulness, Information Quality and Overall Satisfaction is above 4 which indicates that it was appreciated by the students or respondents. Therefore, this system can be a valuable tool for students to manage their finances and make informed decisions about their spending.

The Development of Sensors for Microplastic Detection Using Artificial Intelligence

Bhanuprasad Telu — 2025-12-23

The increasing spread of microplastics throughout the world aquatic ecosystems is a significant ecological and health risk, which highlights an immediate need to develop sophisticated strategies of detection and characterization. The existing analytical approaches to microplastic quantification and identification are commonly not only labor-intensive but also time-consuming and restricted in terms of throughput especially in complicated matrices like soil, river water as well as biosolid fertilizers. Therefore, high-speed, dependable and affordable detection systems are the key to successful environmental surveillance and control measures. To break those limitations, this paper examines the means of integrating artificial intelligence with sophisticated sensor technologies and provides a detailed analysis of the current solutions and suggests new ones to detect microplastic better. In particular, this paper explores the usage of machine learning algorithms to process sensor data, thus making it possible to more efficiently and timely identify, quantify, and even classify microplastic particles. This research paper will seek to give a comprehensive history of some of the sensor modalities, including spectroscopies, optical, and electrochemical techniques, as well as a critical analysis of the AI models, such as deep learning and machine learning, that can be used together to create strong microplastic detection systems. The challenges that this integration tackles include high detection limit, and inability to operate in a portable mode, which is characteristic of the traditional approaches, leading to higher-end, real-time monitoring.