Mastercard (United States)

The rapid development of one of the biggest advances in the sphere of Generative AI has led to natural language generation still being limited to large-scale implementation due to latency, reproducibility, integration complexity, and scalability challenges. This project offers a simple and repeatable deployment pipeline which is based on the Kaggle Fitness Exercises dataset. The final corpus was then fine-tuned with GPT-2 on Hugging Face and PyTorch using extensive preprocessing, that is, data cleaning, instruction merging, normalization, tokenization, and outlier filtering. The model was evaluated on five configurations with a loss of evaluation of approximately 0.4083, 0.4094, and perplexity of 1.5043, 1.5058, and sensitivity and ablation studies indicate the most sensitive hyperparameter is the learning rate. The fine-tuning of the model was performed through a FastAPI service, and it was then containerized using Docker, which made it suitable for deployment on any cloud provider. The outcome was an even training process, low perplexity, and excellent performance reproducibility, while the tests in the production environment showed moderate in inference and throughput, which in turn led to the decision to carry out further optimization on the runtime. The proposed method offers excellent reproducibility, minimum operational cost, and easy deployment compared to the more complex orchestration and full MLOps pipelines.

Retrieval-Augmented Generation (RAG) systems often face limitations in specialized domains such as fintech, where domain-specific ontologies, dense terminology, and acronyms complicate effective retrieval and synthesis. This paper introduces an agentic RAG architecture designed to address these challenges through a modular pipeline of specialised agents. The proposed system supports intelligent query reformulation, iterative sub-query decomposition guided by keyphrase extraction, contextual acronym resolution, and cross-encoder-based context re-ranking. We evaluate our approach against a standard RAG baseline using a curated dataset of 85 question–answer–reference triples derived from an enterprise fintech knowledge base. Experimental results demonstrate that the agentic RAG system outperforms the baseline in retrieval precision and relevance, albeit with increased latency. These findings suggest that structured, multi-agent methodologies offer a promising direction for enhancing retrieval robustness in complex, domain-specific settings.

Deploying reasoning large language models (LLMs) for sentiment analysis requires balancing accuracy, explainability, and efficiency under consumer hardware constraints. We present a cross-architecture evaluation of three reasoning model families-DeepSeek-R1, Qwen3, and GPT-OSS-spanning 17 open-weight variants. Using fine-grained Amazon Reviews sentiment classification with few-shot prompting, we benchmark models on an RTX 4090 Laptop GPU (16 GB VRAM) to assess deployment feasibility. We introduce a novel metric-Reasoning Intensity-to capture trade-offs between explanation depth and computational cost. Results show that GPT-OSS achieves the highest accuracy (86.2 % F1) but exceeds consumer GPU memory limits, while Qwen3 delivers the most practical balance, with the 8B variant reaching 84.8% F1 within 13 GB VRAM. Evaluation on a contemporary benchmark reveals consistent 10-15 percentage point F1 degradation across models, though reasoning intensity remains stable over time. Our findings provide evidence-based guidelines for deploying reasoning-enabled, privacy-preserving sentiment analysis on consumer hardware. Code, datasets, and results will be released to support reproducibility.

Cultural microaggressions pose particular challenges for multinational enterprises due to the vast diversity of employees, customers, and partners across cultures and countries. These subtly yet harmful remarks or actions can lead to negative consequences for firm operations, such as communication disruption, customer and employee attrition, and brand reputational damage. This study investigates the impact of cultural microaggressions and the moderating effects of inclusivity practices and employee coping strategies. Drawing upon the minority stress model, we examine the relationship between cultural microaggressions and various job outcomes. Our findings show a significant effect of cultural microaggressions on job stress with coping strategies attenuating this relationship. We emphasize the importance of multinational corporations fostering more harmonious work environments through inclusiveness practices and equipping employees with coping strategies to mitigate the impact of cultural microaggressions. This study contributes to cross-cultural research by advancing understanding of how subtle behaviors in the form of cultural microaggression shape outcomes across diverse cultural contexts.

Abstract The incredible success in expanding access to financial tools is well documented, with the World Bank’s 2021 Findex reporting that 71% of adults in developing economies now have financial accounts, up from 42% a decade before. This is just the first step. Now governments, nonprofits, and the private sector need to keep working together to encourage greater usage by bringing relevant digital services to more people. These groups should go further by layering in greater connectivity and greater consumer privacy and protections to give users more control over their data. These efforts will help people move up from simply having an account to fully benefiting from it. It’s what we call moving from “financial access” to “financial health.” This work will grow the wave of prosperity even more. This chapter discusses the continued efforts to use digital identity, data analytics, financial education, and many other tools and technologies—as well a network of partnerships—to grow financial health all over the world.

Graph Neural Networks (GNNs), with their ability to effectively handle non-Euclidean data structures, have demonstrated state-of-the-art performance in learning node and graph-level representations. However, GNNs face significant computational overhead due to their message-passing mechanisms, making them impractical for real-time large-scale applications. Recently, Graph-to-MLP (G2M) knowledge distillation has emerged as a promising solution, utilizing MLPs to reduce inference latency. However, existing methods often lack structural awareness (SA), limiting their ability to capture essential graph-specific information. Moreover, some methods require access to large-scale graphs, undermining their scalability. To address these issues, we propose SALE-MLP (Structure-Aware Latent Embeddings for GNN-to-Graph-Free MLP Distillation), a novel graph-free and structure-aware approach that leverages unsupervised structural losses to align the MLP feature space with the underlying graph structure. SALE-MLP does not rely on precomputed GNN embeddings nor require graph during inference, making it efficient for real-world applications. Extensive experiments demonstrate that SALE-MLP outperforms existing G2M methods across tasks and datasets, achieving 3–4% improvement in node classification for inductive settings while maintaining strong transductive performance.

With the rapid digitalization of financial systems and the widespread use of online payment platforms, credit card fraud has become a persistent and escalating threat. Addressing these challenges requires a secure, intelligent, and adaptive detection framework. This study proposes a cloudenabled fraud detection architecture that integrates improved Kepler optimization (IKO) with lightweight cryptography to securely store credit card transaction data in the cloud, ensuring confidentiality and resilience against breaches. For fraud detection, a deep multi-kernel learning integrated with convolutional neural network (CNN) to identify anomalous transaction patterns with high generalization capability. The multi-kernel approach enables the model to capture diverse feature representations across spatial and temporal dimensions of the data. The proposed model was validated using a realworld credit card transaction dataset comprising a high volume of both fraudulent and non-fraudulent records. Experimental results demonstrate the superior performance of our model in terms of accuracy, robustness, and detection efficiency, even under data imbalance conditions. The findings confirm that the integration of CNN with optimized feature selection and secure cloud storage presents optimal solution for real-time credit card fraud identification.

Many analytical tools run off-line and sit in the back room. Someone assembles, formats, and makes sense of the results, who may or may not have anything to do with the decision-making processes. Who sees the results, when, and what they can do with it are pretty much pre-determined. This pre-specified result is then presented to the decision makers who may or may not do anything about it. It is then a one-way communication, feedback is usually not immediate, and it is hard to know what bits and pieces of information are of interest. Interactive tools, on the other hand, allow decision makers to dictate the conditions under which information will be released and in what form. They allow managers to examine different scenarios with relative ease and provide a great deal of scope for tailoring the inputs and the formats required. They enable feedback to be instantaneous and the decision maker is in complete control. In some sense, interactive tools are more like dashboards compared with static analytical models.

As long as financial transactions are recorded and exchanged, it has always been unavoidable that irresponsible actors, or risk factors, engage in various acts of fraud to gain wealth unethically. Such unauthorized transactions include stealing of payment method information, creating unauthorized accounts for receiving gain from illegal activities such as money laundering, creating unauthorized merchant accounts to trick customers into fraudulent transactions, and phishing to swindle customers. Payment processing companies should detect such behaviors to protect legitimate users and save their companies from losing a considerable amount of money (Ghemawat et al., 2003; Kimball & Ross, 2013; Akidau et al., 2015). Growing up in an internet-enriched environment, millennials and Generation Z are the largest active user segment. They are highly sensitive to payment transaction efficiencies and prone to try new payment methods compared to Generation X and Baby Boomers. These new trends require payment service providers to design novel transaction processing systems to accommodate different payment methods, account users, and merchants, while keeping real-time fraud detection at tolerable costs. Acquiring, storing, and processing the intensive data streams generated by users during payment transactions are core challenges for payment processing services. Streaming data engineering is the cornerstone for building such payment processing and risk analysis pipelines. Data such as transactions, partner systems, and payment gateways are collected to a data lake from internal and external sources. Thereafter, the collected data are processed using batch or streaming pipelines to provide real-time transactional and risk insights for the payment operations organizations and partners. Those insights are then ingested into dashboards for tracking transaction activity and risk detection performance.

The digital revolution has profoundly reshaped the landscape of global finance, giving rise to an era where speed, security, and intelligence define the success of financial services. Innovations in Digital Finance and Intelligent Technologies: A Deep Dive into AI, Machine Learning, Cloud Computing, and Big Data in Transforming Global Payments and Financial Services aims to illuminate this transformation by exploring disruptive technologies at the core of today’s financial evolution. In recent years, Artificial Intelligence (AI), Machine Learning (ML), Cloud Computing, and Big Data have moved from theoretical concepts to practical tools that power every facet of modern finance—from algorithmic trading and fraud detection to credit scoring and personalized banking. These innovations are not just enhancing operational efficiency; they are enabling real-time decision-making, improving financial inclusion, and reshaping the way individuals, businesses, and institutions engage with money. This book is crafted for a wide spectrum of readers including financial professionals, technologists, researchers, policy-makers, and students. It unpacks the mechanisms behind intelligent financial systems, provides real-world case studies, and discusses emerging trends such as decentralized finance (DeFi), embedded banking, digital identity, and AI-driven risk assessment. Special attention is given to the regulatory, ethical, and cybersecurity challenges that come with this rapid digitalization. What makes this work timely is its holistic view of innovation—not just as a product of technology, but as a driver of systemic change in global economies. From digital wallets and instant cross-border payments to cloud-native banking platforms, the book outlines how intelligent technologies are setting new standards for agility, transparency, and customer experience. As financial ecosystems grow increasingly complex and interconnected, this book serves as a compass guiding stakeholders toward building secure, inclusive, and intelligent financial futures.