Claude Code is one of the agentic tools that can read, modify, executes and integrate with the IDE, development tools, which helps us to create or fix an issue [1]. The work can be accomplished by reading multiple files and processing the statement to get the right output. Claude code model supports various models, a few are free, and a few are paid service either monthly or yearly. In this research, sample .Net C# console simple application was developed, which would deliver MAPISend [3] and MAPISave[4] email options of the E3 instance developed with the following models: Sonnet-4.6[2], Opus3, Opus 4.6, Opus 4.7, Opus 4.8 [4] and Haiku [10]. The primary purpose is to analyze how each model performs for the same prompt on multiple trials to analysis which is best. It evaluates the efficiency, service offered, usage, correctness and architecture and output generated structure. Results: Finally, this research would help developers to opt for the right model with the default option suggested by Cloud Code, in terms of effectiveness, correctness, time effectiveness, and LOC (lines of code). Claude suggested Sonnet-4.6, comparatively Opus 4.8, based on Bench pro percentage Table 1. Based on real-time evaluation, the research concludes OPUS 4.8. high yields the better results compared to sonnet 4.6.
A swift transformation is being experienced in the electric power sector due to decarbonization, electrification, aging infrastructure, and rising reliability demands. The traditional automation systems used in substations, though useful in monitoring and control, are still highly reactive and silo-based in how they utilize data. New innovations connected to Artificial Intelligence (AI) offer a chance to radically transform substation intelligence and support predictive analytics, autonomous decision-making, and integrated health management of assets. Nevertheless, current deployments are un architecturally coherent, thereby reducing scalability, trust, and operational effects. The proposed paper presents a conceptual and practical architecture of AI-based substations, which combines the concept of Supervisory Control And Data Acquisition (SCADA), grid-edge analytics, machine learning-based intelligence, and governance controls into a single system. The architecture separates substation intelligence into data acquisition, analytics, control, and governance layers, creating effective, auditable, autonomous functionality. The concept of combining the data of protection relays and dissolved gas analysis using AI-based models to reveal the early detection of faults and actionable decision support is illustrated based on an applied reference implementation focused on transformer health intelligence. The paper also analyzes the cybersecurity and governance factors that are critical in the deployment of AI in mission-critical power system settings. The presented framework helps to promote the transition of reactive grid management to resilient, self-healing, and autonomous power systems. The article is concerned with the inadequate architectural solutions to coordinate AI, governance, and grid edge intelligence in contemporary substations.
The study explores a decentralized agentic artificial intelligence orchestration paradigm for enhancing selfrecovery and resilience within a distributed CPS. The traditional paradigms, characterized by their centralized nature, have been observed to present constraints like low scalability, delays in reaction, and susceptibility to single-point failures. As a result, the study has proven to be unsuitable for application in dynamic and complex settings. To solve these problems, a decentralized agentic AI framework is proposed, which utilizes autonomous agents with deep learning models, such as CNN, RNN, and the combined CNN-RNN model for anomaly detection and recovery. According to experimental findings, while the CNN and RNN models exhibit perfect recall performance (1.0000), the study further exhibits high false positives and low reliability scores. Conversely, the hybrid model performs far better than the individual models with regard to F1-score (0.9863), almost perfect AUC score, reduced false positive rates (0.0476), and reliability levels. Despite its relatively high detection latency compared to the individual models, the hybrid framework offers the best trade-off between precision, reliability, and stability.
Bayesian and stochastic models in coronary artery disease extend beyond prediction to clinical implications, including adaptive trial designs, personalized medicine, and enhanced risk stratification. Using a systematic review of literature approach, this study examines Bayesian and stochastic modeling approaches to coronary artery disease progression. The study selects literature using the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) framework, which led to the selection of 17 articles from five databases using some specified inclusion and exclusion criteria. Results show that Bayesian modeling approaches were more prevalent in the predictive landscape of Coronary Artery Disease (CAD), especially when it concerns risk estimation and prognostic modeling. Findings show that while Bayesian techniques focus on probabilistic inference and statistical learning, stochastic techniques prioritize time evolution, physical simulation, and biological plausibility. Results show that Bayesian models perform in individualized risk prediction and calibration, whereas stochastic approaches have strength in providing a deeper understanding of disease dynamics and longitudinal progression patterns. Findings show that Bayesian applications are more patient-specific, while stochastic models support longitudinal and cohort-level management. Results show that Bayesian models are more mature for deployment in predictive clinical environments, while stochastic models provide insight into disease evolution and mechanistic underpinnings. The study establishes that there is a difference between the predictive strength of Bayesian models and the mechanistic interoperability of stochastic models, highlighting the need for integrative modeling frameworks.
Low-resource African languages remain critically underrepresented in natural language processing despite serving hundreds of millions of speakers across diverse linguistic communities. This paper addresses whether explicit morphological feature integration can overcome transformer limitations for Kiswahili, an agglutinative Bantu language spoken by over 100 million people across East and Central Africa. The agglutinative nature of Kiswahili presents fundamental challenges to subword tokenization algorithms that break the grammar and require implicit pattern learning using a small amount of data. The research tested vanilla XLM-RoBERTa on the KenSwQuAD question answering dataset, achieving 20.05% F1 and 17.80% Exact Match on the validation set. This weak baseline performance highlighted the significant limitations of traditional multilingual methods with morphologically complex low-resource languages. The study extended XLM-RoBERTa with explicit representations of 17 Kiswahili morphemes to build a morphologically-enhanced architecture, encoded as multi-hot vectors and introduced through learned projection layers and the pre-trained encoder being frozen to maintain multilingual knowledge. The optimized model's F1 and Exact Match scores of 72.40% and 62.91%, respectively, represented substantial improvements of 52.35 percentage points in F1 and 45.11 percentage points in Exact Match from baseline. Rigorous ablation studies demonstrated that improvements were due to the integration of morphological features, not to model capacity. This work demonstrates that explicit linguistic knowledge integration enables competitive performance even with severely limited training data, providing a reproducible framework for morphologically rich under-resourced African languages and challenging prevailing assumptions about the universal applicability of data-driven approaches.
Artificial Intelligence (AI) has created a pathway to retrieve and calculate data for generating business plans, performing analyses, and conducting research. The value of AI when used in research is intended to reduce the time frame for collecting data and resources. The detriment of new generative AI tools is the question of reliability. This research evaluated two commonly used tools, ChatGPT and Claude, to determine the ability of the tools to recognize (1) false or contradictory data input into the tools directly from the user prompt and (2) false or contradictory data published on the Internet. The results of the study confirm that ChatGPT and Claude have safeguards to prevent data poisoning through direct chat interactions. However, when the bad dataset was published online to a publicly available website, the generative AI tools had difficulty determining validity, introducing concerns for data reliability. This study demonstrates risks regarding the veracity of AI-generated output and its implications for both research and industry.
Financial fraud detection must work extremely fast. Even a small delay of a few milliseconds can affect company revenue and frustrate users. This paper explains how we built a very fast fraud detection system that can handle millions of transactions every second and still make decisions in under 100 milliseconds. The system uses real-time data streaming instead of slow batch processing. It combines event streaming, real-time processing, fast in-memory data storage, and optimized machine-learning models to detect fraud instantly. Compared to traditional batch-based systems, this approach reduces processing time by 73% while still achieving very high accuracy (99.97%). We tested the system on real payment platforms such as digital wallets and peer-to-peer payment systems. According to the test result, the system can process about 2.4 billion transactions per month, with the majority of the decisions completed in under 80 milliseconds, even during peak load. Overall, this work provides both practical design guidance and performance insights for building fast, reliable, real-time fraud detection systems used in critical financial applications.
Today, we are in a rapid technological growth with artificial intelligence, in which privacy and security are the primary concerns. With various language models in the current market, the data that models are trained to provide relevant detail responses plays a vital role in how privacy and security can be leveraged without compromising, for example, the data like Personal Identifiable Information (PI2) [1] and Health Insurance Portability and Accountability ACT (HIPAA) [2]. In this, we provide a review of privacy and security that can be implemented in small industries (specific to the domain) with a Small Language Model (SLM). It also suggests which models are available on the market and how they can be leveraged, considering common factors that align with their business affordability.
The digital divide remains a significant barrier to economic growth, education, and social inclusion, particularly in rural and underserved communities. Despite advancements in telecommunication technologies, millions worldwide lack reliable internet access due to infrastructure limitations, affordability concerns, and digital literacy gaps. This paper presents a data driven approach to bridging the digital divide by leveraging advanced analytics, predictive modeling, and policy-driven solutions. We analyze key factors contributing to connectivity disparities, assess the impact of broadband expansion programs, and propose data-driven strategies for optimizing network deployment and affordability initiatives. The findings offer actionable insights for policymakers, telecommunication companies, and technology providers to accelerate digital inclusion efforts.
The ultimate goal of automating industrial processes is lights-out manufacturing in which manufacturing systems operate independently with either minimal or no human operator involvement. The research paper is a challenge to the technological assumptions, supply chain processes, and strategic requirements that enable the implementation of entirely autonomous factories. It analyzes the role that digitally advanced robotics, AI, the Industrial Internet of Things (IIoT), and digital twins play in the development of self-optimising and resilient manufacturing spaces. The focal attention is put on digital twins as virtual replicas, which are real-time and synchronized, and provide an opportunity to simulate, optimize, and train AI. The paper also outlines an overall operational process that is an overview of the receipt, planning, execution, monitoring, and fulfillment of orders in a lights-out factory. These are material management, autonomous production, in-line quality control, exception management, and inventory management. It is suggested to use an implementation roadmap, covering a gradual development implementation starting with the feasibility analysis, up to the integration of the enterprise system. The study, based on empirical case studies of the industry, points out the success factors and widespread failure modes in lights-out implementations. The article summarizes that, despite the technological choices to consider at any given time, lights-out manufacturing is a method that will be successful when applied in a disciplined way, with strong data-based grounds and a balance between technical independence and the ultimate strategy of the business organizations.
A Question Answering System (QAS) is an Artificial Intelligence (AI) system that automatically understands a user’s question in natural language and provides a precise list of documents that are retrieved. These systems face several difficulties, like understanding context, handling ambiguity in language, and generalization issues. Therefore, a new model, named Autoregressive Secant Optimization-based Retrieval-Augmented Generation Dual View Network (AuRSO-RAGNet), is established for question answering. The proposed model is based on the Taylor-aware Query-RAG Dual View Network with Large Language Model (TQ-RAGNetLLM). The TQ-RAGNet is designed by modifying the Passage Question Answering- Retrieval Augmented Generation (PQA-RAG) approach’s learning rule employing Taylor-Aware Neighbor Mean Square Loss (TANMSL). Answer generation is performed by processing the input passage and questions using PQA-RAG, which comprises two main components: a Question-Dual View Modeling (Q-DVM) and a Question-RAG (Q-RAG) Layer. The overall model is trained using the AuRSO for increasing the performance, where the AuRSO is designed by fusing a Conditional Autoregressive Value-at-Risk (CAViaR) as well as a Secant Optimization Algorithm (SOA). Moreover, experimental outcomes demonstrate that the AuRSO-RAGNet attained a superior recall of 98.248%, precision of 97.840%, and an F1-score of 98.043%.
Blockchain Technology is a Distributed Ledger Technology (DLT) where the data (digital information) is stored in multiple computers and not in a centralized one [1]. Each system would store a copy of the distributed ledger to avoid pitfalls. The information persists as blocks and gets updated simultaneously on all environments after being validated. Four main types of Blockchain, as described: private/permissioned, public/permissionless, hybrid, and consortium [2]. Corda is an open-source platform of a distributed ledger founded by R3 Consortium (R3CEV LLC). DLT is based on peer-to-peer connections with an agreement, and it is not part of the public. Corda architecture is non-native to cryptocurrencies. The platform is based on top of the Java Virtual Machine (JVM), written in Kotlin. Overall, it explains how Corda can be implemented in a wide range of industries with private/permissioned networks. Earlier, blockchain technology was public and permissionless, which posed a little challenge to many industries to adapt, even in the Supply Chain Management system (SCM) and healthcare. Corda is an open-source and DLT concept with private and permissioned features that make it easy to use in industries like SCM, and how that can be achieved.
Cardiovascular diseases continue to be a major cause of mortality worldwide, and among these, cardiac arrhythmias still remain challenging to diagnose accurately. Electrocardiogram (ECG) analysis is a non-invasive tool for the detection of arrhythmias. Manual interpretation of ECG signals can be time-consuming and often varies between clinicians. In this work, six deep learning architectures, ResNet1D-34, InceptionTime, transformer, attention-BiLSTM, attention-aware pooling, and convolution-enhanced transformer, are implemented for beat-level arrhythmia classification using the MIT-BIH arrhythmia database under AAMI class grouping (Normal, Superventricular, Ventricular, Other). All these models are evaluated using macro F1-score to better account for class imbalance, and class-wise evaluation is also assessed to better predict the model performance. The results show that convolution-based models tend to perform more consistently and achieve better balance across classes, while attention-based models struggle with classes that have similar waveform patterns, particularly supraventricular beats.
Effective management of Withholding Tax is vital for ensuring compliance and financial accuracy in modern enterprises operating across multiple jurisdictions. Oracle Fusion Applications address this need through the Fusion Tax module—an integrated, rules-based engine that enables automated tax determination, calculation, and reporting across financial business transactions. This article presents a comprehensive examination of the Withholding Tax framework within Oracle Fusion Tax, detailing its architecture, configuration elements, and business applications. It explores key setup components such as tax regimes, jurisdictions, tax rules, and tax applicability logic, demonstrating how these configurations interact with core financial modules like Accounts Payable. Aimed at implementation partners, business users, and Fusion IT support teams, the article provides both conceptual and practical insights into achieving accurate tax automation, maintaining compliance, and reducing manual intervention. By dissecting the underlying mechanisms and best-practice configurations, this work contributes to a deeper understanding of how Oracle Fusion Tax supports scalable, compliant, and efficient withholding tax management in enterprise environments.
Lung cancer remains one of the leading causes of death worldwide, making early detection important for improving patient outcomes. This study presents a machine learning-based method to predict lung cancer risk by using clinical symptoms along with environmental air quality factors. The system was developed using a dataset of 3000 balanced records containing patient symptoms, Air Quality Index (AQI), and PM2.5 levels. Feature engineering was used to create combined indicators such as smoke anxiety, breath-cough patterns, and pollution exposure. Random Forest and Logistic Regression models were compared, giving accuracies of 56% and 52%, respectively. Although the accuracy is moderate, the results show that environmental factors can contribute to early risk assessment. Among the two models, Random Forest performed better because it captured nonlinear relationships more effectively. This work provides a simple approach that may support preliminary lung cancer risk screening.
To manage assets effectively, companies must also manage how efficiently they operate, minimize the cost involved with doing business, and be able to deliver services consistently and reliably [1]. Many companies depend on disparate systems with limited functionality while still relying heavily on manual processes for maintaining and tracking their asset inventory [2]. This often results in unplanned and excessive legal liability due to niche practices that create opportunities to operate outside the scope of their license agreements, thereby increasing their total costs [3]. In this paper we introduce a comprehensive architecture for building an Enterprise Asset Management (EAM) solution that would integrate the IT Asset Management (ITAM) module of ServiceNow [4], build on a single database to track and manage all assets - including hardware, software, cloud, Internet of Things - and incorporate a fully automated process to orchestrate the entire life cycle of an asset from its creation through to its endpoint. The authors present a case study demonstrating a 25% reduction in downtime of assets for a real-world customer after implementing our solution [8]. A predictive optimization model that utilizes telemetry and predictive machine Learning capabilities [7] is introduced as a way of anticipating asset maintenance requirements and enabling proactive response after identifying the requirement for maintenance. This approach creates an evolving infrastructure enabling improved asset management methodologies well into the future [9].
Artificial Intelligence has moved from reactive systems to systems that can reason, plan, and act over time. This shift has led to the development of Agentic Artificial Intelligence (Agentic AI). Unlike traditional AI systems that follow fixed input output rules, agentic AI systems maintain internal state, reason over context, plan multi-step actions, and interact with external environments with limited human input. This review places agentic AI within the historical development of agent-based artificial intelligence. It describes the main components of agentic systems, including perception, memory, goal management, execution, reflection, and orchestration. It also explains the data infrastructure required to support these systems. Traditional enterprise data platforms and stateless API-based integrations are not designed to support long-term reasoning, continuous context use, or dynamic tool interaction, which limits their suitability for agentic AI. A central focus of the paper is the Model Context Protocol (MCP), which provides a stateful interoperability layer that connects agents with external tools and data sources. MCP is compared with BDI architectures, classical multi-agent systems, and REST-based integration to explain its role and limitations. Security, implementation, and governance issues are also discussed. The review identifies research gaps in coordination, benchmarking, data integration, and secure deployment. Overall, the paper shows that agentic AI builds on earlier agent-based concepts using neural methods and requires layered architectures that combine reasoning models with interoperability standards.
Companies that spend money on enterprise transportation management systems often find that the visibility they expect does not match the technology they have put in place. This paper contends that the issue does not stem from the platforms themselves, which have undergone substantial enhancement, but rather from the organizational conditions that determine the effective utilization of these platforms. Based on a structured review of the literature on supply chain management and evidence from the implementation of SAP Transportation Management (SAP TM) in the SAP S/4HANA environment, this study presents the Logistics Visibility Maturity (LVM) framework. The LVM framework is a diagnostic tool that measures an organization's readiness in three areas: transportation planning transparency, shipment execution monitoring, and logistics analytics usage. The LVM framework shows a four-level maturity progression that helps organizations figure out where they are now, what gaps are keeping them from full visibility, and how to make targeted improvements. The results show that the most important change is not going from having no system to having one, but from just having a system to using it regularly. This is a detail that is often missed in current implementation guides. The framework offers a theoretical advancement in supply chain visibility research and a pragmatic approach for logistics organizations implementing TMS.
Cancer causes a large number of deaths around the world every year. To diagnose cancers correctly, doctors examine tissue images carefully, but doing this manually takes a lot of time, and different doctors can reach different conclusions from the same image. This study presents a deep learning model that combines several techniques, such as ResNet18, Vision Transformer (ViT), Graph Convolutional Network (GCN), and Graph Attention Network (GAT), to classify these cancer images more accurately. ResNet18 is used to capture detailed local features from the images, while ViT analyzes the global context by understanding how different parts of the image relate to each other. GCN and GAT further model and refine the structural relationships between features. The novelty of this work lies in integrating convolutional, transformer-based, and graph-based learning within a single framework to jointly capture local, global, and relational information for histopathological image classification. Experimental evaluation on the LC25000 dataset demonstrates that the proposed ViT+GAT architecture achieves improved classification accuracy and generalization performance compared to standalone ResNet18, ViT, and GCN models. These results indicate that the proposed approach can support more reliable and efficient automated cancer diagnosis in computational pathology.
Mobile Edge Computing (MEC) has become increasingly critical for latency-sensitive applications, including Augmented/Extended Reality (AR/XR), Cloud Gaming, Real-Time Video Analytics, and Interactive Enterprise Services. Existing edge steering mechanisms remain largely reactive by relying on static policies, nearest-edge selection, or compute-only information that usually fail under user mobility, fluctuating radio conditions, dynamic user-plane paths, and edge resource contention rather than being more proactive. This paper presents SCONE-AEGIS framework that extends the Standard Communication with Network Elements (SCONE) paradigm beyond throughput advisories to support joint network-compute steering of MEC applications. SCONE-AEGIS introduces an Edge Steering Advice (ESA) that communicates recommendations that can be consumed by the applications, which have been derived from a combination of RAN, UPF, and MEC telemetry. The framework is a combination of a two-stage AI/ML engine, the first being a Spatio-Temporal Graph Predictor that is uncertaintyaware and models the evolving relationships among radio access network nodes, user-plane functions, edge sites, and mobile users, and the second stage is a Safe Contextual Bandit Steering Policy (SCBSP) that selects execution sites subject to SLA constraints, migration hysteresis, and prediction confidence. The proposed framework provides a standards-compatible, privacypreserving path for exposing joint network-compute intelligence to applications without breaking transport encryption.
