Volume 14 | Issue 7 |

 Abstract

Critical infrastructure systems -power grids, water treatment facilities, and fuel distribution networks -face an unprecedented wave of sophisticated cyber-physical attacks that simultaneously compromise digital control networks and manipulate physical sensor telemetry to evade detection. Between 2019 and 2022, recorded attacks on industrial control systems surged by 140%, with incidents such as the Oldsmar water treatment intrusion and the Colonial Pipeline ransomware attack exposing the catastrophic cost of undetected adversarial access. Existing defences fail for two structural reasons: signature-based intrusion detection systems cannot identify zero-day or composite attack patterns, while centralised machine learning approaches require raw operational data to leave infrastructure premises, violating data sovereignty mandates.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Urban waste management is increasingly facing systemic challenges due to rapid urbanization, population growth, and rising consumption levels. Several studies highlight that existing waste management systems remain fragmented, relying heavily on manual data collection, delayed reporting, and centralized databases that lack transparency and efficiency. Research on blockchain- based waste management solutions demonstrates improved traceability and secure record keeping; however, most existing implementations focus only on tracking waste flows rather than enabling intelligent decision-making.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Multi-Agent System; Blockchain; IoT; Smart City; Waste Management; Random Forest; Route Optimization; Smart Contracts; Ethereum

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The goal of precision oncology is to use patient-specific data to personalize cancer treatment; however, current AI-based systems have poor interpretability, little personalization, and no safety constraints. The Digital Twin-Driven Safe and Explainable Meta-Deep Reinforcement Learning (DT-SMDRL) framework for adaptive therapy optimization is proposed in this paper. A transformer-based fusion model is used to integrate multimodal patient data, such as multi- omics, clinical records, imaging, and physiological signals, into a single representation. In a risk-free setting, a patient-specific digital twin mimics the course of the illness and the results of treatment. Through knowledge transfer between patient populations, a meta-deep reinforcement learning agent based on proximal policy optimization facilitates quick personalization. The framework uses risk-sensitive learning and safety-constrained optimization with toxicity limits to guarantee clinical reliability. Treatment reasoning is improved by causal modelling, and decisions are made with confidence thanks to uncertainty estimation. Interpretability at the biomarker level is provided by an explainable AI module. Continuous learning from actual results is made possible by a closed-loop feedback mechanism. The framework is appropriate for safe and adaptive precision oncology because experimental results show increased treatment efficacy, decreased toxicity risk, quicker adaptation, and improved interpretability.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Precision Oncology, Digital Twin, Meta-Reinforcement Learning, Safe Reinforcement Learning, Explainable AI, Causal Modeling, Uncertainty Estimation, Multimodal Data Fusion

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The rapid growth of digital transaction systems has increased the demand for secure and reliable user authentication mechanisms. Traditional methods such as passwords and PINs are highly vulnerable to phishing, credential thefts, and unauthorized access, making them unreliable for modern security requirements. Although facial recognition has emerged as a convenient biometric alternative, it remains vulnerable to spoofing attacks using photographs, videos, or replayed media.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Facial Recognition, Liveness Detection, AWS Rekognition, Digital Security, Biometric Authentication, Multi-Factor Authentication Anti Spoofing.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Phishing remains a pervasive cybersecurity threat, often bypassing traditional defenses that operate reactively only after a user interacts with a malicious link. To address this limitation, this paper presents ZeroClick Defender, a proactive, hybrid phishing detection framework. Crucially, our approach introduces hover-based, pre-click detection powered by millisecond-scale local inference, identifying threats before any user action occurs. The system achieves this through a multi-layer edge-cloud architecture. Initially, a lightweight local model provides rapid heuristic filtering and risk estimation within 10 milliseconds to ensure a seamless user experience. For complex or uncertain cases, the system seamlessly defers to a deep-scan cloud pipeline that performs advanced structural and linguistic analysis, resolving threats in approximately one second. By combining fast edge-side processing with robust cloud analysis, ZeroClick Defender achieves high accuracy against novel phishing URLs, providing a reliable, scalable, and privacy-friendly defense against modern social engineering attacks.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Phishing Detection, Machine Learning, Cloud Computing, Heuristic Rules, Client-Cloud Architecture.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Mechanical workshop training in engineering curricula confronts fundamental structural constraints: insufficient machine availability, elevated material expenditure, occupational safety hazards, and limited instructor capacity. While hands-on engagement remains essential, students encounter restricted equipment access, substantial waste generation, and injury risks during formative skill acquisition. This paper presents an extended reality (XR) system designed to serve as a preparatory and complementary modality to conventional hands-on instruction, enhancing skill acquisition velocity while decreasing material consumption and facilitating consequence-free procedural exploration. We introduce a design-oriented framework comprising four integrated dimensions: physics-accurate operation simulation encompassing lathe, milling, and drilling machinery; competency evaluation via operation-specific performance indicators aligned with institutional workshop benchmarks; graded difficulty progression reflecting established learning pathways; and safety scenario simulation that normalizes hazard awareness and risk mitigation. The framework emphasizes process-level accuracy and mechanism behavior replication rather than photorealistic rendering, facilitating economically viable institutional deployment. This analysis encompasses implementation methodology, financial viability, economic modeling, and integration tactics, demonstrating anticipated break-even within 18-24 months via material consumption reduction and instructor workload optimization. Our methodology establishes XR as a complementary preparatory tool that enhances student competency and safety consciousness prior to physical workshop engagement.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Extended Reality (XR), Virtual Reality, Mechanical Training, Physics-Based Simulation, Competency Assessment, Safety Training, Engineering Education, Immersive Learning.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Cloud-native back-ends fail open in a particular and uncomfortable way. Once a user has cleared the login form and a JWT has been minted, the request path stops asking questions. Role-Based Access Control (RBAC) treats every subsequent call as equally trustworthy, even when the behaviour around it has shifted in ways an operator would notice instantly. This paper describes ZTaaS, an AI-driven Zero Trust Security-as-a- Service platform built around an external reverse-proxy gateway that re-evaluates trust on every hop. ZTaaS combines a Node.js gateway that performs JWKS-based RS256 verification, runtime policy evaluation and short-lived internal token translation; a continuous telemetry pipeline that captures behavioural features into MongoDB and recomputes per-tenant baselines on a scheduled job; a low-latency deviation-based risk score in the request path paired with an asynchronous Isolation Forest model running off a RabbitMQ queue; and an adaptive enforcement layer that maps the resulting score onto allow, step-up, or block actions without touching the protected back-end. We describe the multi-window feature extraction (60 s, 10 min, 60 min) used to keep freshly authenticated users out of the false-positive bucket, the weighted deviation score, and the unsupervised model bootstrapped on a synthetic baseline of normal traffic with a small number of injected anomalies. The implementation runs in Node.js, Python and React. End-to-end tests show that authentication and authorisation can be enforced as a single source of truth at the edge, that high-risk sessions are short-circuited at the proxy boundary, and that the analytics fan-out adds no measurable latency to the request path under nominal load.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Zero Trust Architecture, Security-as-a-Service, API Gateway, Continuous Authentication, Anomaly Detection, Isolation Forest, Cloud-Native Security, Policy-as-Code

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Voice communications in enterprises face several difficulties in terms of scalability, intelligent call routing, and real-time analytics. In this paper, an intelligent cloud-based multi-agent system that leverages distributed computing, natural language processing (NLP), and serverless computing is proposed. The multi-agent system coordinates multiple autonomous agents responsible for intelligent call routing, voice transcriptions, chatbot-based conversations, and insight creation. All these processes occur simultaneously via WebSocket-based communication channels. The multi-agent system is implemented on the Amazon Web Services platform, where the functions are deployed using the Lambda service and EC2 machines. The implementation results show a marked improvement in call routing, voice transcriptions, and operational insights generated by the agents.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Multi-Agent Systems, Serverless Computing, Real-Time Communication, Intelligent Call Routing, Speech-to-Text Systems, Natural Language Processing, Distributed Cloud Systems, WebSocket Protocol, Conversational Agents, Voice Analytics

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Conventional document notarization is based on centralized systems with added cost, delay and single points of failure. This paper introduces a decentralized notarization system that uses blockchain, IPFS, and new web technologies to offer verifications of digital documents that are not interchangeable, transparent and maintain privacy. The system creates a hash of the uploaded documents in the SHA-256 format, safely stores the file in IPFS and documents the hash with a time rating on an ethereum-based smart contract. Staking and reputation are used to encourage honest behavior among the notaries, and full-stack architecture can ensure that it is easy and scalable. Extensive testing of smart contracts, backend services, and frontend elements was very reliable and well functional. The real-world feasibility was tested by being deployed on the Sepolia testnet. The presented solution avoids the use of the centralized mediators and provides integrity, availability, and user-controlled certification of the digital records.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Decentralized notarization, Blockchain, Smart contracts, IPFS, Document integrity, Ethereum, Distributed systems.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Generative AI's development to the point where synthetic media can be so close to actual media that it raises issues; e.g. when people use manipulated video clips of public figures to spread false information (deepfakes), they create confusion regarding truthfulness and decrease reliability for all forms of electronic information. This research explains the development of a method to deal with this issue both effectively and practically, by constructing an automated system that not only detects deepfakes with high accuracy, but also provides an explainable reason(s) as to why it categorizes specific items as deepfakes. This method also includes the use of machine learning-based algorithmic feature extraction for building the overall framework and machine learning algorithms for determining the explainable results. The overall framework and method have a very defined process from data collection, isolation of facial areas, preprocessing images and finding features of the images via convolutional neural networks and then classifying the images. During this process, the model will focus on finding those minute discrepancies (slight distortions in texture and unnatural blending) that are the signatures of deepfake models even if they are convincing to people. The features that are extracted during the feature extraction step are then compressed to form a smaller, more manageable representation and sent to the classifier to determine if the image is real or manipulated; however, just using accuracy was not enough for us and we also added an explainability layer by using Grad-CAM so that we could view which areas the model is focusing on. Sometimes it is the eyes, sometimes it is the jawline and, interestingly, it is not always what a human would look at first. Ultimately, the goal of this entire project is to not only detect deepfakes, but to understand why they are detected as deepfakes so that we can develop systems that will be able to eliminate the possibility of deepfakes in the future.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Deepfake Detection, Explainable AI, Grad-CAM, SHAP, LIME, CNN, Transformer, InceptionV3, Xception, CSWin Transformer.

  License

Creative Commons Attribution 4.0 and The Open Definition