Information Technology and Security

Multi-cloud strategy adoption has introduced significant architectural fragmentation across enterprise platform engineering teams. While organizations leverage multiple cloud providers for resilience and vendor lock-in avoidance, divergent identity models, networking abstractions, governance mechanisms, and deployment constructs create operational complexity and inconsistent security postures. Existing literature emphasizes cloud comparison and workload portability, with limited focus on systematic platform engineering harmonization. This article presents a governance-driven enterprise multi-cloud platform harmonization framework standardizing platform engineering practices across heterogeneous cloud environments. The framework addresses architectural abstraction alignment, identity federation normalization, network segmentation consistency, and deployment pipeline portability. Implementation across Kubernetes-based platforms in Amazon Web Services and Microsoft Azure incorporated Infrastructure-as-Code automation, identity equivalence mapping, and policy-driven governance controls. The framework introduces the Deployment Consistency Index, Identity Alignment Score, and Governance Conformance Ratio to quantify standardization outcomes. Empirical evaluation demonstrates reduced configuration variance, improved deployment reproducibility, minimized cloud-specific policy exceptions, and enhanced governance traceability. This work contributes a reusable architectural blueprint for multi-cloud platform standardization with empirical evidence supporting governance-centered harmonization strategies.

A multifaceted framework integrating personalized employee training programs with behavioral analysis techniques is proposed in this paper to help mitigate insider threats in organizations. Based on a comprehensive literature review and empirical evaluation, employees are portrayed here as a potential vulnerability but also as a critical component of the organizational defense system. The proposed framework can easily be scaled across organizations of different sizes and sectors, with practical guidelines provided for how it might be implemented in the real world. The investigation of the integration of sophisticated technologies into the behavioral analysis process provides for the creation of adaptive and personalized training. The results of the study highlight the ability of this framework to enhance organizational resilience in a dynamic cybersecurity threat landscape.

Large Language Models (LLMs) have transformed natural language processing applications, but their capabilities remain highly dependent on how users formulate their inputs. This paper presents a comprehensive analysis of prompt engineering strategies across multiple dimensions, including model types, task categories, and domain applications. Through systematic experimentation with different prompt structures, context optimization techniques, and task-specific approaches, we quantify the performance impact of various prompting paradigms. Our findings reveal that Chain of Thought and Step-by-Step strategies consistently outperform other approaches, with an average performance improvement of 14.8% across tasks. We also demonstrate that optimal prompt structure varies significantly by model architecture and intended application, with context window utilization showing diminishing returns beyond 70%. Based on our experiments, we propose a unified framework for prompt design that adapts to model capabilities, task requirements, and resource constraints. Our contribution advances the understanding of how prompt engineering impacts LLM performance and provides actionable guidelines for practitioners to maximize model utility across diverse applications.

The accelerating sophistication of web-based threats has outpaced the capacity of purely automated or exclusively human-centered defense postures. This paper examines the integration of human expertise with machine intelligence as a necessary evolution in web threat response systems. Through a synthesis of recent scholarship in explainable artificial intelligence, cognitive security, and human-automation interaction, the study proposes a conceptual framework wherein machine learning models handle volumetric detection and pattern recognition while human analysts exercise judgment over ambiguous, novel, or high-impact threats. Key findings indicate that successful integration depends upon explainability mechanisms that make algorithmic decisions transparent, trust calibration protocols that prevent automation bias or disuse, and role differentiation that respects the complementary strengths of human and artificial cognition. The analysis also identifies persistent challenges, including the vulnerability of human-AI teams to adversarial manipulation and the organizational inertia that impedes workflow redesign. This paper concludes that effective web threat response resides not in replacing human operators but in reconfiguring their relationship with intelligent systems a shift from automation to genuine augmentation.

The extended complexity of the electronics control units (ECUs) of autonomous and electric cars makes it necessary to implement fault-tolerant designs that comply with the ISO 26262 ASIL-D. The paper will discuss how hardware-software co-design is used in guaranteeing the safety and reliability of automotive sensor ECUs. The systematic review of 21 articles published between 2021 and 2025 lists integrated strategies related to redundancy, virtualisation, artificial intelligence, and cybersecurity to attain the fail-operational resilience. In the research, the codesigned systems have been shown to have a 90 per cent diagnostic coverage, less than 5 ms recovery latency, and 95 per cent fault detection performance, which is much better than the traditional modular design. Hardware redundancy ensures physical resilience, and adaptive software enables the tasks and proactive fault recovery to be transmitted without difficulties. Moreover, there are cybersecurity features, including voltage-based ECU fingerprinting and root-of-trust verification, to improve the reliability of communications. This paper suggests the Co-Design Reliability Enhancement Framework (CREF) that has the capability of guaranteeing compliance with ASIL-D through the incorporation of redundancy, artificial intelligence, and fault prediction, as well as pipeline testing. The framework illustrates that cybersecurity and functional safety will need to go together, and the ideas of co-design underlie the design of the next-generation, software-defined, fault-tolerant vehicles.

Mission-critical data centers typically rely on generator sets that operate for a small percentage of their total lifecycle. Accurately calibrating these generator sets is crucial to meeting operational reliability and efficiency while reducing replacement costs and maximizing asset lifetime. However, conventional methods do not integrate lifecycle data from the generator sets or the synchronization of calibration and control. This work proposes a digital twinassisted predictive calibration solution that incorporates sensor streams from generator sets, telemetry from a preventive maintenance system, associated weather predictions, and the demand profile of Artificial Intelligence-supported workloads running on the data centers. The research concludes by analysing the robustness of the prediction of the generator-set parameters when the response of the Data Centre is exposed to variation from the normal workload profile for which the Data Centre was designed.

Mission-critical data centers typically rely on generator sets that operate for a small percentage of their total lifecycle. Accurately calibrating these generator sets is crucial to meeting operational reliability and efficiency while reducing replacement costs and maximizing asset lifetime. However, conventional methods do not integrate lifecycle data from the generator sets or the synchronization of calibration and control. This work proposes a digital twinassisted predictive calibration solution that incorporates sensor streams from generator sets, telemetry from a preventive maintenance system, associated weather predictions, and the demand profile of Artificial Intelligence-supported workloads running on the data centers. The research concludes by analysing the robustness of the prediction of the generator-set parameters when the response of the Data Centre is exposed to variation from the normal workload profile for which the Data Centre was designed.

Generative artificial intelligence (GenAI)-particularly large language model (LLM) tools-has rapidly entered educational practice and is beginning to reshape science teaching, learning, and assessment. Science education is a distinctive use case because it requires epistemic reliability: learners must justify claims with evidence, apply disciplinary constraints (e.g., units, conservation laws), and engage in inquiry practices. This chapter offers a PRISMA 2020-aligned systematic review of recent research on GenAI in science education, complemented with a policy and ecosystem analysis for India. Evidence from published systematic reviews and empirical studies suggests that GenAI can support explanation, scientific writing, formative feedback, and inquiry planning when embedded in well-designed tasks. However, risks persist: hallucinations and inaccuracies, bias, privacy concerns, and academic integrity threats, especially where institutional guidance is limited. In India, NEP 2020 and NCF-SE 2023 emphasize competency-based learning, technology integration, and scientific temper, while national digital infrastructure (DIKSHA and NDEAR) provides a scalable platform for teacher professional development and content delivery. This chapter synthesizes evidence into an India-ready implementation framework (S-CIENTIFIC), proposes assessment redesign options, and provides classroom-ready prompt templates, rubrics, and a reproducible search strategy (databases and Boolean strings).

The increasing reliance on large-scale distributed digital infrastructures in banking systems and national security environments has intensified the need for advanced forensic intelligence capabilities capable of operating in Big Data ecosystems. This paper proposes a scalable Big Data–driven forensic intelligence framework integrating distributed data processing, real-time streaming analytics, machine learning–based anomaly detection, and structured forensic decision systems for the protection of critical infrastructures. The framework operates across heterogeneous data sources including financial transaction systems, network telemetry, cloud logs, IoT devices, and governmental information systems. By combining forensic science principles with Big Data architectures, the proposed approach enables proactive detection, reconstruction, and attribution of cyber incidents affecting financial stability and national security.                 

References List
1. Guarino, A. Digital Forensics as a Big Data Challenge. Springer eBooks 2013, 197–203.

2. Mpungu, C.; George, C.; Mapp, G. Digital Forensics Readiness in Big Data Networks: A Novel Framework and Incident Response Script for Linux–Hadoop Environments. Applied System Innovation 2024, 7 (5), 90.

3. Song, J.; Jin, L. A Framework for Digital Forensic Investigation of Big Data. 2020.

4. Huy, S.; Ang, S. H.; Ho, M.-X.; Balasubramaniam, V. Insider Threats in Banking Sector: Detection, Prevention, and Mitigation. 2025, 2025 (4), 257–265.

5. Gilbert, C.; Gilbert, M. A.; Dorgbefu, M. Detection and Response Strategies for Advanced Persistent Threats (APTs). 2025, 5–21.

6. Danang, D.; Mustofa, Z.; Irlon, I. Digital Forensics and Automated Incident Response Framework Leveraging Big Data Analytics and Real Time Network Traffic Profiling in Heterogeneous Cyber Environments. 2026, 1 (1), 45–55.

7. Ekundayo, F. Big Data and Machine Learning in Digital Forensics: Predictive Technology for Proactive Crime Prevention. World Journal of Advanced Research and Reviews 2024, 24 (2), 2692–2709.

8. Suhartono, D.; Burhandenny, A. E.; Wardani, R.; Isnaini, K. N.; Hidayat, T. Forensics in the IoT Era: Navigating Challenges and Designing Future Approaches. 2025, 364–369.

9. Real-Time Financial Data Processing Using Apache Spark and Kafka. International journal of data science and machine learning. 2025, 05 (01), 137–169.

10. Trambadiya, D. C. From Streams to Security: Architecting a Production-Ready Fraud Pipeline with Flink and Kafka. Zenodo (CERN European Organization for Nuclear Research) 2026.

11. Event Driven Fraud Detection Pipeline: Real-Time Processing with Kafka, KsqlDB  Apache Flink. International Journal of Computer Applications 2025, 187 (60), 13–18.

In today's digital age, the importance of security cannot be overstated. With the rise of cyber threats, businesses and organizations face constant danger from hackers, malware, and other malicious activities. Traditional security measures, while still relevant, often struggle to keep up with the speed and complexity of these threats. This is where machine learning (ML) comes into play, significantly enhancing real-time threat detection.

Traditional vendor risk management relies on fixed calendar schedules for reassessments, creating fundamental misalignment between predetermined review cycles and continuously evolving risk landscapes. This article presents a systematic signal-driven framework that translates heterogeneous continuous monitoring signals across several risk domains, security posture, privacy and data governance, operational resilience, enterprise risk management controls, financial viability, sanctions and financial crime, and reputational indicators, into standardized risk metrics suitable for threshold-based reassessment triggering. The framework employs statistical normalization techniques, including z-score analysis, rate-of-change calculations, and severity scoring, to convert diverse monitoring events into comparable Control Impact Scores, which aggregate through weighted summation calibrated to vendorservice characteristics. Inherent risk levels modulate control effectiveness changes to project residual risk movement, with reassessment triggers activating when risk changes exceed governance-defined materiality thresholds within signal-appropriate drift windows. Empirical calibration through historical back-testing optimizes precision and recall while episode-based correlation analysis identifies compounded exposures across multiple degrading domains. Comprehensive audit trail documentation transforms algorithmic triggers into transparent governance decisions supporting regulatory examination, while quarterly threshold review committees enable adaptive refinement based on observed risk and performance metrics. The framework provides organizations with defensible, risk-intelligent reassessment timing that responds to actual vendor risk trajectory rather than arbitrary schedules, focusing review resources where genuine control degradation occurs while maintaining appropriate oversight across the vendor portfolio.

Deepfake technologies-encompassing generative adversarial networks, diffusion-based synthesis, and transformerdriven voice cloning-have evolved from entertainment novelties into precision socio-technical weapons targeting organizational trust infrastructures. Existing countermeasures focus narrowly on artifact-level detection accuracy, failing to model systemic vulnerabilities within enterprise decision chains, financial authorization workflows, and executive identity trust graphs. This paper introduces the Cognitive Authenticity Assurance Architecture, a novel multilayered defense framework integrating Cognitive Attack Surface Modeling, Trust Graph Disruption Index analytics, Adversarial Co-Evolution Defense Engine, Authenticity-by-Design Protocol with cryptographic watermarking, and Zero-Trust Media Verification Architecture. The framework reconceptualizes deepfake threats as cognitive supply-chain attacks on organizational trust ecosystems rather than isolated media forgeries. A graph-theoretic Trust Graph Disruption Index metric quantifies synthetic identity propagation risk across enterprise communication networks. Simulation results demonstrate a 47% reduction in successful executive impersonation attacks, 39% improvement in synthetic media identification speed within financial workflows, and 52% reduction in decision-layer compromise probability. This architecture advances deepfake mitigation beyond detection into organizational trust engineering, establishing a new operational and theoretical paradigm for enterprise cognitive resilience.

This research presents a comprehensive conceptual framework for the
application of Artificial Intelligence (AI) in Smart City cybersecurity,
specifically designed for the year 2030. The study addresses the critical
"Reaction Gap" in current security models by proposing the Parallel
Agentic Architecture, a system that replaces traditional manual triage
with a synchronized swarm of specialized AI agents (Scout, Analyst,
and Responder). This framework is built upon theoretical modeling
and system-level analysis, introducing pioneering concepts such as the
“Glasswing Protocol” for autonomous self-hardening and the
identification of “Stop-hook Violations” within digital supply chains.
The core of the research explores the synergy between high-speed
autonomous defense and transparent human governance. By utilizing
Tiered Model Routing (ranging from high-efficiency tiers like Haiku to
high-reasoning tiers like Claude Mythos), the architecture ensures both
economic sustainability and technical invincibility. A significant focus
is placed on accountability; the study integrates OpenTelemetry 2.0
and the “Mind Replay” mechanism to provide verifiable reasoning
traces for every AI decision, ensuring full compliance with the AI
Integrity Act.
Furthermore, the research evaluates the evolving role of the human
engineer, transitioning from reactive troubleshooting to the role of the
AI Policy Architect (AIPA). This professional manages the “Synergy
Coefficient” between machine velocity and human ethical judgment to
resolve logic deadlocks between competing urban systems (e.g., Traffic
vs. Energy). The findings conclude that a “Single-Layer Experience”
for end-users, combined with community-centric resilience, is essential
for a sustainable urban shield. Ultimately, this work provides a
roadmap for securing the smart cities of the future while preserving
political freedom, social connectivity, and the safety of every citizen.

The rise of Large Language Models (LLMs) has created new era for AI driven cybersecurity. As cyber threats become increasingly sophisticated, the need for autonomous intelligent systems to identify vulnerabilities, analyze attack pattern and manage risk in real-time increases. LLMs can enhance vulnerability detection, malware analysis, intrusion detection, phishing detection, and policy interpretation. However, they lack the autonomous decision-making and action capabilities required for real-time proactive enforcement. Agentic AI, by creating a topology of LLM-based reasoning with autonomous decision-making and action taking capabilities, overcome these limitations.
Zero-Trust Architecture (ZTA) enforces strict access controls by default, operating on the principle that no entity, whether users, IoT devices, or services should be trusted implicitly, instead, access to resources is
granted on a per-request basis, ensuring that only necessary actions are permitted and lateral movement is minimized.
This research re-architect Zero Trust as a self organizing multi-agent system, where each zero trust function is enforced by the autonomous AI agents. Agentic-ZTA framework integrates autonomous contextual agents,
which provide fine-grained access control, real-time threat detection, and dynamic policy adaptation with symbolic reasoning. In addition, a trust algorithm aggregates input from specialized agents to create a dynamic trust score, to resolve conflicting decisions. Our proposed autonomous Agentic AI based Zero Trust architecture (Agentic-ZTA) explores an innovative solution to modern cybersecurity challenges and demonstrates its utility by providing a fully integrated agent coordination and automated response for zero trust enforcement in a dynamic contested environment.

Modern organizations are considerate of enterprise data and AI platforms, which usually provoke excessive and unpredictable operating expenses. Dynamism in cost management Traditional cost management systems use manual reviews, and reports that are backwards looking and which are never effective when dealing with dynamic clouds. The paper is suggesting a Telemetry-Driven Cost Governance model which operates on a real-time platform telemetry, automated policy assessment and controlled execution in order to govern costs on a continuous basis. The model combines the compute, storage, and data pipeline telemetry into a single analytics layer and measures the costs against specific preconceived thresholds. The approval gates and safety checks are automated through recommendations that are carried out in order to maintain system stability. A quantitative analysis presents good outcomes. The total platform cost has been cut by 24 percent, and compute cost is lowered by 28 percent and storage cost has been cut down by 19 percent. Effort of manual analysis of costs was reduced 90%. There was an increase in policy compliance of 76-96 and job success rate was 97.8 percent to 98.6 percent respectively. The findings have shown that Telemetry-based governance can provide a proactive, safe, and quantifiable cost optimization of enterprise-level data and AI platforms.

Data mining of enterprise data is also changing to include natural language processing (NLP) and machine learning (ML) to derive insights of structured business intelligence systems. The present paper will suggest an intelligent data mining system of SAP BusinessObjects and SAP Business Warehouse (BW) with the help of the NLP-enabled model that utilizes the state-of-the-art language models that will be implemented on Amazon Web Services (AWS). The framework enables its users to issue natural language queries to find and process enterprise data and eliminates the divide between non-technical decision-makers and big SAP data repositories. Our approach is based on a model based on BERT that is fine-tuned to vocabulary of enterprise domains and an AWS cloud-based system to transform queries to data warehouse functions. Experiments show that information retrieval is highly accurate and that the query processing time is much reduced as compared to the traditional methods. Our findings indicate that our solution has 95 percent accuracy when presented with queries on a sample with a 40 percent processing speed increase. The paper has indicated a way in which NLP and cloud-based AI can be used together to transform business analytics that would allow organizations to view data more intuitively, quicker in decision-making, and provide more insights.

The lakehouse architecture provides a promising foundation for future-proofing enterprise data platforms. Empirical research examined Microsoft Fabric Lakehouse as a real-world case. Findings confirm AI-ready capabilities across the core components and architecture. Building block services, including data import and data preparation, enable AI readiness in scale. However, the capacity of a single organization to drive trust and privacy controls, lineage, metadata, auditability, and compliance with regulatory frameworks is also crucial. These characteristics are typically defined and governed by enterprise-wide ecosystems that responsibly share Microsoft Fabric Lakehouse resources with other data and analytical infrastructures. Élite Computing has established hybrid cloud scenarios with Microsoft Azure, enhanced with an automatic training engine using DataRobot to predict not only a customer's next purchase but when might their risk of churn happen or which products are likely to be purchased together. AI readiness has been explored and discussed in heterogeneous computing environments such as PLCs and SCADA establishing robust enterprise data platforms and meeting performance criteria. However, verifying the enterprise operation of these AI-ready capabilities in AI workloads covering the complete workflow-from data preparation and feature engineering to model training, deployment, and monitoring-represents an important next step.

Hybrid cloud environments offer organizations essential flexibility and scalability by integrating private and public infrastructure, but this complexity introduces the core research problem: a greatly expanded attack surface and inconsistent application of security controls, leading to critical vulnerabilities and data exposure risks. The study employed a mixed-methods design, analyzing extensive secondary data through quantitative findings combined with qualitative thematic analysis of security reports. The key results quantified the main contributors to data leakage, finding misconfigured APIs as the leading threat (35%), followed by endpoint vulnerabilities (30%), and encryption failures (25%). While efficient AES-256 encryption is common, the research noted critical challenges in key management and the continued use of outdated protocols. Furthermore, inadequate use of secure API management frameworks was highlighted as significantly increasing the risk of unauthorized access. The implications of these findings led to the development of a project management framework that addresses these gaps. This framework integrates cloud-specific solutions-including end-to-end encryption, secure API configurations, automated compliance monitoring, and advanced endpoint detection and response (EDR) tools-within a PRINCE2 method. This proposed multi-layer approach provides proactive risk management, effectively bridging architectural gaps and improving the overall security posture against technical and compliance issues across the hybrid environment.

Traditional access control models built on role-based permissions and direct database queries fail when applied to AI-mediated data access in regulated industries. AI creates an intermediary layer that breaks fundamental assumptions: users don't query databases directly, AI tools do; authorization decisions made at build time cannot account for runtime context; and audit trails designed for human access patterns don't capture AI behavior. This paper presents a three-layer architecture that evaluates access control at runtime rather than at build time, enabling compliant AI deployment within environments governed by HIPAA, SOC2, GDPR, and similar frameworks.

The rapid expansion of cloud-centric enterprise architecture and large-scale remote work deployments has positioned Virtual Private Network (VPN) technology at the center of contemporary cybersecurity practice. Among available VPN paradigms, Internet Protocol Security (IPSec) and Secure Sockets Layer VPN now technically implemented via Transport Layer Security (TLS) represent two architecturally distinct solutions to the common problem of securing communications across untrusted public networks. Although both achieve confidentiality, integrity, and authentication, they diverge substantially in their operational layer within the OSI model, the breadth of traffic they protect, their performance characteristics under enterprise load conditions, and their suitability across different deployment scenarios. This paper presents a comprehensive, individually conducted comparative analysis of IPSec and SSL VPN across three dimensions: security posture, encompassing encryption cipher suites, authentication mechanisms, and vulnerability surface; performance characteristics, including throughput efficiency and protocol-induced latency under varying load conditions; and scalability potential, addressing enterprise deployment complexity, cloud readiness, and integration with emerging Zero Trust Network Access (ZTNA) and Secure Access Service Edge (SASE) frameworks. Through analytical modelling grounded in published protocol specifications and peer-reviewed benchmarking literature, this study finds that IPSec maintains measurable advantages in throughput and latency at high loads, while SSL VPN provides superior accessibility, granular access control, and cloud-native scalability. Recommendations for protocol selection across enterprises, remote-access, and cloud scenarios are presented alongside directions for future research, including post-quantum cryptographic migration.

Multi-Factor Authentication (MFA) has emerged as a critical cybersecurity mechanism for protecting digital systems against unauthorized access. Traditional authentication methods relying solely on passwords are increasingly vulnerable to cyber threats such as phishing, credential stuffing, and data breaches. MFA enhances security by requiring users to provide two or more independent verification factors, including knowledge-based, possession-based, and biometric credentials. This research explores the concept, operational mechanisms, applications, and relevance of MFA within modern cybersecurity frameworks. It highlights the effectiveness of MFA in reducing security breaches while addressing its limitations and future trends such as adaptive authentication and passwordless systems.

The modern information environment has radically changed with the accelerating expansion of high-speed data streams of Internet of Things devices, digital interactions, and financial systems. The classic batch processing processes become more and more deficient in the current enterprise demands that require the ability to make instant decisions in milliseconds after the data is generated. Although Lambda architecture is meant to provide a tradeoff between fault tolerance and low-latency processing by having separation of batch and speed layers, it induces significant complexity in operation by having two sets of codebase maintenance and code duplication. The Kappa architecture is a conceptual simplification that assumes all data are continuous streams and removes an architectural duplication of Lambda architectures, in addition to making substantial complexity and resource reduction infrastructural. Stateful computations with many operators can be executed on modern stream computing infrastructures that provide advanced state management schemes to sustain stateful computations such as temporal joins, pattern matching, and long-running aggregations that tolerate operator failure. Layered storage architectures make the distinction between hot, warm, and cold layers of storage data through the use of access patterns and latency requirements, and cloud services use advanced tiering strategies where data can be automatically moved across the storage classes. Column-based storage systems are more effective in analytical queries with a better compression ratio and lower input-output needs, and multi-level caching solutions greatly minimize unnecessary calculations. The event-driven architectures encourage loose coupling among the system components by using event production and consumption patterns that change the flow of information radically, in contrast to the normal request-response systems. The integration of integrated stream processing, efficient state management, and optimized storage architecture provisions allows organizations to satisfy their scalability requirements and enhance maintainability, resource utilization, and democratize real-time analytical performance across a wide range of organizational environments.

March 9, 2026
Today I submitted a formal audit request to federal authorities regarding Medicaid billing for Applied Behavior Analysis (ABA) services in California. This follows a pilot investigation I conducted in San Diego examining how autism “therapy” services are billed to Medi-Cal.

The full report is now public: https://doogri.org/2026/03/09/cms-audit-request-san-diego-pilot-march-6-2026/

For months I have been analyzing how the ABA industry bills Medicaid. What I found raises serious program-integrity concerns that deserve federal scrutiny.

The authors in this paper present an observability system that is robust enough and able to fit in real-time payment systems and compliant-aware mechanisms fit into operational reliability and regulation compliance. The framework enables systems to respond more effectively to perturbations such as network partitioning, latency amplification, and cyber-attacks, using intelligent monitoring, anomaly detection and security-focused observability pipelines. In experimental testing, both in synthetic and real-world deployment scenarios, Mean Time to Detect (MTTD) is reduced, incident recoverability is increased, and compliance alerting is optimized. With the help of case studies and empirical measurements, we confirm the strength of observability-based resilience in strengthening digital financial infrastructures. The result offers practical information to financial organizations aiming at achieving high-availability, secure, and regulatory compliant payment systems.

Invoices are a difficult task to automatically extract key information because of the variability of invoice layouts and prohibitive cost of manual annotation, though this step is vital to automating financial  workflows.  This  paper  introduces  a  template-free  lightweight  understanding framework  of  invoices  that  works  efficiently  without  depending  either  on  layout-specific principles  or  highly  annotated  datasets.  In  contrast  to  the  traditional  systems  based  on handcrafted  templates  or  computationally  expensive  models,  the  suggested  solution  takes advantage of the vision-language reasoning of SmolVLM and the use of the heuristic-based post-processing in order to identify the necessary fields within a broad variety of invoice templates including invoice number, date, vendor name, VAT, tax, discount, sub-total and total amount, etc. The experimental findings show that the framework is robust in respect to a wide range of multi-layout invoices, providing a viable and scalable solution to the automation by virtue ofinvoice in the real-world. Transparency and auditability are also highlighted in the proposed method  which  allows  extracted  information  to  be  simply  validated  in  compliance-driven domains  like  banking  and  finance.  In  addition,  due  to  its  lightweight  nature,  it  has  low computational  overhead  and  is  hence  deployable  in  both  large-scale  and  even  in  small  to medium-sized businesses with low resource capabilities. The framework integrates semantic reasoning and rule-based validation thus bringing forward the development of explainable and resource-efficient intelligent document processing, placing it as an alternative to template-based or transformer-intensive models in practice

The pervasive nature of internet memes in digital culture has led to their increasing adoption in marketing. While anecdotal evidence suggests their effectiveness, a comprehensive academic understanding of how meme-based content digital marketing affects consumer behavior remains underexplored. This study addresses this gap by employing the Stimulus-Organism-Response (SOR) model to investigate the psychological processes through which meme qualities translate into consumer responses. Objectives: The primary objective of this study is to investigate the effectiveness of meme-based content digital marketing by examining the relationships between meme qualities (content appeal, emotional triggers, and relevance), consumers' internal states (cognitive processing, emotional reaction, and brand perception), and their behavioral responses (shareability, engagement, and purchase intention). The study aims to empirically confirm the SOR model as a theoretical framework for understanding the psychological processes underpinning the effectiveness of meme-based content marketing through digital platforms. Methods: A quantitative research design was employed utilizing the partial least squares structural equation modeling (PLS-SEM). Data were collected from social media users who had encountered meme-based digital marketing content. A researcher-made questionnaire with a seven-point Likert scale was used to measure perceptions on meme quality, internal state, and behavioral response. Results: Descriptive statistics indicated high mean scores across all constructs, suggesting that meme-based content digital marketing is effective in eliciting favorable perceptions on meme qualities, consumers' internal states, and behavioral responses. The measurement model demonstrated satisfactory reliability and validity. Structural model analysis revealed a significant positive influence of Meme Quality on Internal State (β=0.577, t=5.183, p<0.001) and Internal State on Behavioral Response (β=0.663, t=5.741, p<0.001). Crucially, the Internal State significantly mediated the relationship between Meme Quality and Behavioral Response (βmediat=0.383, t=4.185, p<0.001). However, the direct path from Meme Quality to Behavioral Response was not statistically significant (β=0.135, t=1.269, p=0.205). Conclusions: This study confirms the effectiveness of meme-based content digital marketing, operating primarily by shaping consumers' internal states. The findings strongly support the applicability of the SOR model, demonstrating that while meme qualities do not directly lead to behavioral responses, they significantly influence consumers' cognitive processing, emotional reactions, and brand perceptions, which in turn drive shareability, engagement, and purchase intention. Digital marketers should focus on creating appealing, emotionally engaging, and relevant memes to cultivate these positive internal states.

This systematic review aims to understand what research has been done on the use of AI as a classroom tutor and how that body of work should shape future research. A systematic review was conducted using key term searches in four major, peer-reviewed journal databases for relevant research and categorized them by research type, AI type, assistance type, and education level and analyzed them for study measures and key findings. After applying inclusion and exclusion criteria, a total of 34 research articles were analyzed for key findings. The review found that most studies have focused on chatbots supporting undergraduate level education. Key findings show that AI tutors may be most useful for students beginning at a lower level of expertise than their peers, and the use of techniques such as RAG may significantly enhance AI's usefulness as a tutor with niche and higherlevel subjects. This systematic review uniquely considers the last decade of research on AI use as a classroom tutor and provides insights into where future research on the subject may be most useful and impactful to the field of education.

The rapid proliferation of synthetic media, presentation attacks, and document forgeries has created significant vulnerabilities in Know Your Customer (KYC) workflows across financial services, telecommunications, and digital-identity ecosystems. Traditional monolithic KYC systems lack the scalability and agility required to counter adaptive fraud. This paper proposes an Agentic AI Microservice Framework that integrates modular vision models, liveness assessment, deepfake detection, OCR-based document forensics, multimodal identity linking, and a policydriven risk engine. The system leverages autonomous micro-agents for task decomposition, pipeline orchestration, dynamic retries, and human-in-the-loop escalation. Experimental evaluations demonstrate improved detection accuracy, reduced latency, and enhanced resilience against adversarial inputs. The framework offers a scalable blueprint for regulated industries seeking robust, real-time, and privacy-preserving KYC verification.

Modern enterprises deploying applications across heterogeneous cloud platforms face critical security challenges in protecting data transmitted during continuous integration and deployment workflows. Traditional single-layer encryption mechanisms prove insufficient for multi-tenant cloud environments where data traverses organizational boundaries, network perimeters, and jurisdictional zones during automated build and release processes. This paper presents a novel cryptographically layered security framework implementing automated dual-encryption protocols within continuous deployment pipelines spanning Microsoft Azure and Google Cloud Platform infrastructures. The proposed architecture employs Advanced Encryption Standard 256- bit symmetric encryption as primary protection layer with Rivest-Shamir-Adleman 4096- bit asymmetric encryption providing secondary security envelope, both managed through hardware-backed key management systems integrated directly into build automation workflows. Implementation leverages Python-based orchestration scripts executing within Azure DevOps pipelines to generate ephemeral cryptographic material, enforce zero-trust access policies, and maintain comprehensive audit trails satisfying General Data Protection Regulation and Health Insurance Portability and Accountability Act compliance requirements. Empirical evaluation across production deployment scenarios demonstrates the framework achieves 89.4 percent reduction in theoretical breach exposure while introducing only 12.3 percent overhead in pipeline execution time. The system successfully processes enterprise workloads exceeding 50 terabytes monthly across geographic regions while maintaining sub-second key rotation capabilities and automated compliance reporting. This work establishes a reproducible methodology for securing inter-cloud communications in regulated industries requiring cryptographic protection beyond conventional transport layer security implementations

This study investigates the role of Generative AI (GenAI) in creating Teaching and Learning Resources (TLRs) and enhancing the pedagogic skills of basic school teachers in Ghana's Upper East Region. Despite the Ministry of Communication, Digital Technology, and Innovations