Amazon Web Services Research Articles

Background Alzheimer’s disease and related dementias (ADRD) affect nearly five million older adults in the United States, yet more than half remain undiagnosed. Speech-based natural language processing (NLP) provides a scalable approach to identify early cognitive decline by detecting subtle linguistic markers that may precede clinical diagnosis. Objective This study aims to develop and evaluate a speech-based screening pipeline that integrates transformer-based embeddings with handcrafted linguistic features, incorporates synthetic augmentation using large language models (LLMs), and benchmarks unimodal and multimodal LLM classifiers. External validation was performed to assess generalizability to an MCI-only cohort. Methods Transcripts were obtained from the ADReSSo 2021 benchmark dataset ( n = 237; derived from the Pitt Corpus, DementiaBank) and the DementiaBank Delaware corpus ( n = 205; clinically diagnosed mild cognitive impairment [MCI] vs. controls). Audio was automatically transcribed using Amazon Web Services Transcribe (general model). Ten transformer models were evaluated under three fine-tuning strategies. A late-fusion model combined embeddings from the best-performing transformer with 110 linguistically derived features. Five LLMs (LLaMA-8B/70B, MedAlpaca-7B, Ministral-8B, GPT-4o) were fine-tuned to generate label-conditioned synthetic speech for data augmentation. Three multimodal LLMs (GPT-4o, Qwen-Omni, Phi-4) were tested in zero-shot and fine-tuned settings. Results On the ADReSSo dataset, the fusion model achieved an F1-score of 83.32 (AUC = 89.48), outperforming both transformer-only and linguistic-only baselines. Augmentation with MedAlpaca-7B synthetic speech improved performance to F1 = 85.65 at 2 × scale, whereas higher augmentation volumes reduced gains. Fine-tuning improved unimodal LLM classifiers (e.g., MedAlpaca-7B, F1 = 47.73 → 78.69), while multimodal models demonstrated lower performance (Phi-4 = 71.59; GPT-4o omni = 67.57). On the Delaware corpus, the pipeline generalized to an MCI-only cohort, with the fusion model plus 1 × MedAlpaca-7B augmentation achieving F1 = 72.82 (AUC = 69.57). Conclusion Integrating transformer embeddings with handcrafted linguistic features enhances ADRD detection from speech. Distributionally aligned LLM-generated narratives provide effective but bounded augmentation, while current multimodal models remain limited. Crucially, validation on the Delaware corpus demonstrates that the proposed pipeline generalizes to early-stage impairment, supporting its potential as a scalable approach for clinically relevant early screening. All codes for LLMCARE are publicly available at: GitHub .

Most datasets suffer from partial or complete missing values, which has downstream limitations on the available models on which to test the data and on any statistical inferences that can be made from the data. Several imputation techniques have been designed to replace missing data with stand in values. The various approaches have implications for calculating clinical scores, model building and model testing. The work showcased here supports using an Item Response Theory (IRT) based approach for categorical imputation, comparing it against several methodologies currently used in the machine learning field including k-nearest neighbors (kNN), multiple imputed chained equations (MICE) and Amazon Web Services (AWS) deep learning method, DataWig. Analyses comparing these techniques were performed on three different datasets that represented ordinal, nominal and binary categories. The data were modified so that they also varied on both the proportion of data missing and the systematization of the missing data. Two different assessments of performance were conducted: accuracy in reproducing the missing values, and predictive performance using the imputed data. Results demonstrated that the proposed method, Item Response Theory for categorical imputation, fared quite well compared to currently used multiple imputation methods, outperforming several of them in many conditions. Given the theoretical basis for the approach, and the unique generation of probabilistic terms for determining category belonging for missing cells, IRT for categorical imputation offers a viable alternative to current approaches.

Amazon Web Services (AWS) Elastic Compute Cloud (EC2) instances offer scalable computing resources crucial for various applications. Accurate prediction of CPU utilization is essential for efficient resource management and cost optimization in cloud environments. This study investigates the performance of machine learning models, specifically Long Short-Term Memory (LSTM) networks and AutoRegressive Integrated Moving Average (ARIMA) models, for forecasting CPU utilization of AWS EC2 instances in both development and production environments. By employing historical data from both environments, the research aims to extend predictive horizons and improve forecasting accuracy. We evaluate and compare model performance using Mean Squared Error (MSE) and fitting times. Results reveal that ARIMA models consistently outperform LSTM models in terms of MSE and computational efficiency, demonstrating superior performance in both environments. LSTM models, despite their potential, show higher variability and longer fitting times, especially with hyperparameter tuning. This paper highlights the critical role of model selection and tuning in enhancing forecasting capabilities and operational efficiency in cloud resource management. The findings contribute valuable insights for optimizing resource allocation and cost management in AWS cloud services.

Cloud computing has become essential for businesses, providing scalable, flexible, and cost-efficient solutions. As organizations increasingly rely on cloud-based infrastructure, selecting the most suitable provider for specific needs is crucial. This research evaluates four leading cloud platforms—SiteGround, Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP)—to determine the optimal platform for hosting the TAKDA WebCPaSMS proto-model. Using a comparative case study design and content analysis, the study examines provider reliability, security, scalability, cost-effectiveness, and technical support. Findings indicate that SiteGround is ideal for managed cloud hosting and small-scale applications due to its ease of use but lacks traditional compute, storage, and networking services. AWS provides the most comprehensive cloud solutions, Azure integrates best with Microsoft enterprise ecosystems, and GCP excels in AI, data analytics, and cloud-native applications. These insights help organizations select the best cloud platform for their needs.

ABSTRACT Patient demographic prediction involves estimating age, gender, ethnicity, and other personal characteristics using X‐rays. This can help in personalized medicine and improved healthcare outcomes. It can assist in automated diagnosis for some diseases that exhibit age and gender‐specific prevalence. It can also help in forensic science to identify individuals when demographic information is missing. Insights from deep learning can verify the gender and age of self‐reported individuals through chest X‐rays (CXRs). In this proposed work, we have deployed an artificial intelligence (AI) enabled model which focuses on two tasks: gender classification and age prediction from CXRs. For gender classification, the model combines ResNet‐50 (CNN) and Vision Transformer (ViT) to leverage both local feature extraction and global contextual understanding for predicting gender and is called ViTCXRResNet. The model was trained and validated on an Amazon Web Services (SPR) dataset of 10702 images, split with an 80–20 ratio, which was evaluated with classification metrics to determine the model's behavior. For age prediction, extracted features from ResNet‐50 were used with dimensionality reduction through principal component analysis (PCA). A fully connected feedforward neural network was trained on the reduced feature set to predict age. The classification and regression model achieves accuracy results of 93.46% for gender classification and 0.86 for the R 2 score for age prediction on the SPR dataset. For visual interpretation, explainable AI (Gradient‐weighted Class Activation Mapping) was utilized to visualize and find out which parts of the image are prioritized for classifying gender. The proposed model yields high classification accuracy in gender detection and significant accuracy in age prediction. The model shows competitive accuracy compared to existing methods. Further, the demographic prediction stability of the model was proven on two different ethnic groups, such as the Japanese Society of Radiological Technology (JSRT) and Montgomery (USA) datasets.

This whitepaper presents a comprehensive analysis of multi-cloud data synchronization patterns for Microsoft Dynamics 365 Finance and Operations (D365 F&O) integrated with Amazon Web Services (AWS) and Google Cloud Platform (GCP). As organizations increasingly adopt multi-cloud strategies to optimize performance, reduce vendor lock-in, and enhance disaster recovery capabilities, the complexity of maintaining data consistency across distributed cloud environments has become a critical challenge. Our research examines four primary synchronization patterns: real-time synchronization using OData APIs and Business Events, batch processing through the Data Management Framework (DMF), event-driven architectures leveraging cloud-native messaging services, and hybrid approaches that intelligently route data based on criticality and volume requirements. Through analysis of implementation case studies across e-commerce, manufacturing, and financial services sectors, we demonstrate that hybrid synchronization patterns achieve optimal performance with 99.9% data consistency while reducing operational costs by up to 40%. Key findings include: (1) Real-time patterns excel for transactional data with sub-second latency requirements but face API throttling limitations of 200 requests per minute; (2) Batch processing patterns efficiently handle high-volume data transfers but introduce latency measured in hours; (3) Event-driven patterns provide superior fault tolerance and scalability through loose coupling; and (4) Hybrid patterns optimize resource utilization by matching synchronization method to data characteristics. This research contributes to the body of knowledge on enterprise multi-cloud integration by providing practical implementation guidance, performance benchmarks, and security frameworks that organizations can leverage to design robust, scalable, and secure multi-cloud data synchronization architectures.

Abstract Background and Aims While evidence indicate that prolonged intradialytic hypoxemia is associated with increased morbidity and mortality in HD patients (Meyring–Wösten et al., CJASN 2016.), research into arterial oxygen saturation (SaO2) patterns during hemodialysis (HD) is scarce. Here we aim to describe the intradialytic dynamics of SaO2 in a large and diverse cohort of HD patients with arteriovenous vascular access. Understanding the SaO2 dynamics may open new avenues to identify HD patients at high risk for adverse clinical outcomes. Method SaO2 levels during HD were measured using the Crit-Line monitor (CLM; Fresenius Medical Care, Waltham, MA). CLM measurements were recorded at 10-second intervals and transmitted into Amazon Web Services (AWS) via Apache KAFKA, a real time streaming software. We extracted CLM data between January 2021 and July 2023. A sample data collected on January 2, 2023, was analyzed. During data cleaning, recordings with SaO2 ≤0 or ≥99.9% and hematocrit <15% or >60% were deemed implausible and hence excluded. We then excluded sessions with an SaO2 standard deviation of <0.1%, an SaO2 interquartile range of zero, and less than a total of 60 minutes of recordings. Also, sessions with less than 5 minutes of measurements either at the start or at the end of the session were filtered. Mean SaO2 was calculated for each time point (every 10 seconds) and a plot using locally estimated scatterplot smoothing (LOESS) regression with shaded 95% confidence intervals was generated to analyze the trend of SaO2 during treatments. Results The per-10-second average and the LOESS function of SaO2 from 6,068 HD sessions from 6,068 patients are shown in Fig. 1. Right after HD started, SaO2 declined and reached its nadir around 60 minutes into the treatment. Thereafter, SaO2 increased and finally reached levels close to starting values. Conclusion Our results of a biphasic intradialytic SaO2 dynamic corroborate an earlier report (Campos et al. Blood Purif 2016). The decline of SaO2 in the first hour on dialysis is notable. We consider two potential explanations. First, the flux of bicarbonate from the dialysate may increase blood pH and reduce the ventilatory drive. Second, the early nadir in SaO2 may be associated with transient, subclinical bio-incompatibility of components in the extracorporeal dialysis circuit that may result in leukostasis in small pulmonary vessels and reduced gas exchange. We think that the subsequent increase in SaO2 may be attributed to ultrafiltration, which alleviates pulmonary congestion and improves oxygen exchange. Whether and to what extent these observations correlate with adverse clinical outcomes warrants further investigation. Real time analysis of continuously measured physiological parameters may provide valuable insights into dialysis pathophysiology.

The Industrial Internet of Things (IIoT) revolutionize industries such as manufacturing, logistics, energy, and healthcare by merging smart sensors and devices with sophisticated network connectivity and advanced data analysis. Digital Twin As a Service (DTaaS) for Internet of Healthcare Things (IoHT) in the healthcare industry opens exciting opportunities to create virtual replicas of real healthcare systems and assets. Digital twins relying on cloud platforms, such as Amazon Web Services, Microsoft Azure, and Google Cloud, provide several vital capabilities, including data informed decision making, personalized patient simulation, up to 30% decrease in equipment downtime with predictive maintenance, and operational efficiency of more than 25% through real-time remote monitoring. This paper proposes an all encompassing methodology towards the development and deployment of compositional digital twins utilizing services applied towards assisting the visually challenged with a smart stick. Federated learning has been proposed as one potential approach that could help in preserving the privacy of clients, particularly concerning the protection of patient's confidential information. One of the possible healthcare scenario that demonstrates how digital twin technology guiding visually impaired individuals, with a possible enhancement in the success rate of mobility by 40%.

Enterprise IT has been transformed due to cloud computing, but organizations have never stopped wrestling with maximizing performance, cost, and scalability. Amazon Web Services has been the first to incorporate the application of artificial intelligence and machine learning technologies to solve these issues on its platform. The article performs an in-depth analysis that analyzes how AWS uses AI to improve three essential aspects that including autoscaling desktops, serverless computing frameworks, and workload optimization approaches. AWS is reinventing legacy resource management strategies using such ideas as Predictive Scaling, Compute Optimizer, Karpenter, Lambda cold start optimization, and DevOps Guru, which are creating smarter, self-optimizing systems. The technologies are used to analyze historical and real-time data, which can be used to make predictive decisions, identify optimization opportunities, and automatically adjust resources according to fluctuating demands. In the article, it is revealed how these AI-intensified services bring meaningful advantages in the form of resource utilization, application performance, operational efficiency, and environmental sustainability. They are witnessing the evolution of such systems towards cross-service optimization, sustainability intelligence, intent-based computing, and fleet-wide learning, thus presenting a paradigm shift in cloud infrastructure management that is redefining how firms look at their cloud computing strategy.

Cloud adoption is now more of a multi-vendor solution rather than a single vendor solution, and multicloud solutions, whereby organizations have taken the services of more than one vendor to gain flexibility, resiliency, and performance. Oracle has now established a foothold in this spot with interoperability between the Oracle Cloud Infrastructure (OCI), Amazon Web Services (AWS), and Microsoft Azure. Oracle workloads can be run directly on these platforms through projects such as Oracle Database@Azure and Oracle Database@AWS, as well as dedicated interconnect services, thereby reducing latency and avoiding vendor lock-in. The current paper covers the opportunities and challenges of Oracle multicloud interoperability. It investigates building patterns, performance benefits, costs, and risks in a qualitative study based on technical documentation and case studies. The findings show that the Oracle strategy has the potential to enhance the portability of the workloads, disaster recovery, and agility. It requires a prudent approach to governance and integration practices. The article sheds light on the contribution of Oracle to changing the enterprise IT strategy and offers advice to the association that has to work in multifaceted multicloud settings

Misconfigured cloud infrastructure has emerged as a prevalent and impactful threat vector in cybersecurity. In particular, organizations deploying services on Amazon Web Services (AWS) often face significant security risks due to incorrectly configured access controls, insufficient logging, weak network segmentation, and the absence of critical protections such as Web Application Firewalls (WAF). Despite the widespread adoption of Infrastructure as Code (IaC) tools (e.g., Terraform, AWS CloudFormation) to enforce predictable, version-controlled deployments, these IaC configurations typically undergo little to no systematic security testing. Unlike application code—which routinely undergoes unit, integration, and security testing—infrastructure code is seldom tested beyond basic static analysis or post-deployment monitoring. As a result, critical security misconfigurations can remain undetected until they are exploited by attackers. To address this gap, this paper proposes a novel approach termed "Infrastructure as Tested Code." By applying proven software testing techniques—such as test assertions and continuous integration workflows—to IaC, our framework enables pre-deployment validation of an AWS environment’s security posture. We develop and evaluate a proof-of-concept implementation that automates security checks for Terraform-defined AWS resources, focusing on key configurations including WAF rules, Identity and Access Management (IAM) policies, S3 bucket permissions, and security group rules. This test suite is built with open-source tools (Chef InSpec and AWS SDKs) and runs within a CI/CD pipeline using LocalStack to emulate AWS services. Through this approach, developers and DevSecOps teams can detect and remediate misconfigurations early in the development lifecycle, long before infrastructure reaches a production environment. Our experimental evaluation shows that integrating automated security tests into the DevOps pipeline significantly strengthens cloud security and mitigates misconfiguration-driven vulnerabilities. Compared to traditional static analysis tools, our approach offers greater flexibility, supports environment-specific policies, and allows developers to codify custom, testable security assertions. Even a minimal test suite proved effective in catching high-risk misconfigurations that static checks overlooked. This paradigm complements existing cloud security tools (such as AWS Config, Checkov, and other policy-as-code frameworks) and can be seamlessly integrated into DevSecOps pipelines. Finally, the paper provides a detailed implementation guide, a real-world case study, and an analysis of practical trade-offs. We conclude that just as test-driven development improved software reliability, adopting a test-driven approach to infrastructure can become a critical strategy for proactively securing cloud environments. This work lays the groundwork for future research into formalizing security testing practices for IaC, benchmarking IaC security test coverage, and developing reusable libraries of security test assertions for AWS.

This article is dedicated to the development and research of an advanced hybrid machine learning method for time series forecasting in decision support systems (DSS). The relevance of the work is driven by the rapid growth of data volumes in modern information systems, particularly in cloud infrastructures, and the need for accurate forecasting tools for effective resource management. The objective of the study is to increase the accuracy of computing resource load forecasting by developing a hybrid model that combines the advantages of statistical methods and deep learning architectures. A novel hybrid architecture is proposed, integrating the Autoregressive Integrated Moving Average (ARIMA) model for modeling linear components of a time series, and a Long Short-Term Memory (LSTM) recurrent neural network with a built-in Attention Mechanism for analyzing non-linear residuals. The ARIMA model is used to capture stationary dependencies and seasonality, while the LSTM network with an attention mechanism effectively models complex, non-linear, and long-term patterns in the data remaining after ARIMA processing. An experimental study was conducted on a real dataset of CPU utilization monitoring from virtual machines in the AWS (Amazon Web Services) cloud environment. The proposed hybrid ARIMA-LSTM model with an attention mechanism demonstrated a significant improvement in forecasting accuracy compared to baseline models: pure ARIMA, pure LSTM, and a standard hybrid ARIMA-LSTM model. The Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) metrics for the developed model were 12-18% lower than those of the best-performing baseline model. Scientific novelty lies in the enhancement of existing hybrid approaches by integrating an attention mechanism into the LSTM architecture for analyzing time series residuals. Practical significance of the work consists in the potential for implementing the developed method in automated DSS for optimizing resource allocation, auto-scaling, preventing overloads, and reducing operational costs of cloud infrastructure.

The monitoring and analysis of public clouds is gaining momentum due to their widespread exploitation by individual users, researchers, and companies for daily tasks. This paper proposes an algorithm for optimizing the cost and utilization of a set of running Amazon EC2 instances by resizing them appropriately. The algorithm, named Cost and Utilization Optimization (CUO) algorithm, receives information regarding the current set of instances used (their number, type, utilization) and proposes a new set of instances for serving the same load, so as to minimize cost and maximize utilization, or increase performance efficiency. CUO is integrated into Smart cloud Monitoring (SuMo), an open-source tool developed by the authors for collecting and analyzing monitoring data from Amazon Web Services (AWS). A number of experiments are performed using input data that correspond to realistic AWS configuration scenarios, which exhibit the benefits of the CUO algorithm.

BackgroundIn 2019, the Open Pediatric Brain Tumor Atlas (OpenPBTA) was created as a global, collaborative open-science initiative to genomically characterize 1,074 pediatric brain tumors and 22 patient-derived cell lines. Here, we present an extension of the OpenPBTA called the Open Pediatric Cancer (OpenPedCan) Project, a harmonized open-source multiomic dataset from 6,112 pediatric cancer patients with 7,096 tumor events across more than 100 histologies. Combined with RNA sequencing (RNA-seq) from the Genotype-Tissue Expression and The Cancer Genome Atlas projects, OpenPedCan contains nearly 48,000 total biospecimens (24,002 tumor and 23,893 normal specimens).FindingsWe utilized Gabriella Miller Kids First workflows to harmonize whole-genome sequencing (WGS), whole exome sequencing (WXS), RNA-seq, and Targeted Sequencing datasets to include somatic SNVs, indels, copy number variants, structural variants, RNA expression, fusions, and splice variants. We integrated summarized Clinical Proteomic Tumor Analysis Consortium whole-cell proteomics and phospho-proteomics data and miRNA sequencing data, as well as developed a methylation array harmonization workflow to include m-values, beta-values, and copy number calls. OpenPedCan contains reproducible, dockerized workflows in GitHub, CAVATICA, and Amazon Web Services (AWS) to deliver harmonized and processed data from over 60 scalable modules, which can be leveraged both locally and on AWS. The processed data are released in a versioned manner and accessible through CAVATICA or AWS S3 download (from GitHub) and queryable through PedcBioPortal and the National Cancer Institute’s pediatric Molecular Targets Platform. Notably, we have expanded Pediatric Brain Tumor Atlas molecular subtyping to include methylation information to align with the World Health Organization 2021 Central Nervous System Tumor classifications, allowing us to create research-grade integrated diagnoses for these tumors.ConclusionsOpenPedCan data and its reproducible analysis module framework are openly available and can be utilized and/or adapted by researchers to accelerate discovery, validation, and clinical translation.

Multi-cloud database architectures represent a strategic evolution in enterprise data management, enabling organizations to leverage the strengths of multiple cloud providers while mitigating risks associated with vendor dependency and single points of failure. This article examines the implementation of Oracle Database and SQL Server technologies across distributed cloud environments, focusing on the technical frameworks, architectural patterns, and operational strategies that enable successful multi-cloud deployments. The article explores critical components, including Oracle Data Guard and GoldenGate replication mechanisms, SQL Server Always On Availability Groups, and cross-cloud integration strategies that facilitate seamless data distribution and high availability across Amazon Web Services and Microsoft Azure platforms. Key implementation challenges are addressed, including data latency optimization, network connectivity considerations, security architecture consistency, and identity management synchronization across heterogeneous cloud environments. The article evaluates management and orchestration tools such as Azure Arc and AWS Outposts, while examining real-world case studies from enterprise migrations, financial services implementations, manufacturing data distribution models, and healthcare compliance scenarios. Performance analysis frameworks, cost-benefit considerations, and reliability measurements provide quantitative insights into the comparative advantages of multi-cloud versus single-cloud deployments. Future directions encompass emerging trends in serverless database technologies, artificial intelligence integration, edge computing architectures, and quantum-safe security implementations that will shape the evolution of multi-cloud database strategies. The article demonstrates that while multi-cloud database architectures offer substantial benefits in terms of operational resilience, vendor flexibility, and strategic positioning, successful implementations require comprehensive planning, sophisticated technical expertise, and robust governance frameworks to address the inherent complexities of distributed cloud environments.

The increasing demand for efficient and scalable loan management systems has prompted the need for a modernized approach that leverages cloud technology. This research focuses on the design and development of a cloud-based Loan Management System (LMS) aimed at enhancing the process of loan application, approval, disbursement, and repayment management. Traditional loan systems often suffer from inefficiencies, limited scalability, and security risks, which this system seeks to address through cloud computing. The system is built using cloud infrastructure, ensuring high availability, security, and costeffectiveness. Key features of the system include automated loan application processing, real-time approval workflows, dynamic interest rate calculation, repayment tracking, and user role management for loan officers and borrowers. The use of cloud-based technologies such as Amazon Web Services (AWS) facilitates a flexible, scalable, and secure environment for storing sensitive data and ensuring data integrity. Through an iterative development methodology, the system was tested for functionality, performance, and security. Functional testing confirmed that the LMS meets the core requirements of the loan management process. Performance testing demonstrated the system’s scalability under different loads, and security tests ensured that user data is securely handled through encryption and strong authentication mechanisms. This system offers several advantages over traditional on-premises solutions, including reduced infrastructure costs, ease of scaling, and real-time data access. It is anticipated that this cloud-based LMS will be beneficial for financial institutions seeking to modernize their loan management processes and provide an improved user experience for both administrators and borrowers.

The National Institute of General Medical Sciences (NIGMS) Sandbox, in collaboration with the NIH Office of Data Science Strategy (ODSS), NIH Center for Information Technology (CIT), and Deloitte, has developed an innovative, cloud-based training module on protein crystallography. This module, designed to enhance accessibility and scalability, is hosted on Amazon Web Services (AWS) and is part of a broader initiative to provide free, interactive training in biomedical data science. The protein crystallography module is divided into three submodules, each focusing on a critical aspect of the crystallographic process: (1) an introduction to protein structure and crystallographic theory, (2) solving structures using single-wavelength anomalous diffraction (SAD) phasing, and (3) model building and refinement using molecular replacement. The first submodule provides a foundational understanding of protein structure, emphasizing the relationship between structure and function, and introduces key concepts in crystallography, including crystallization strategies and the phase problem. The second submodule delves into SAD phasing, guiding users through the process of solving protein structures using anomalous diffraction data. The third submodule focuses on model building and refinement, utilizing the Crystallographic Object-Oriented Toolkit (Coot) and the Phenix software suite to teach students how to interpret electron density maps, build protein models, and refine structures for deposition into the Protein Data Bank (PDB). Each submodule integrates Jupyter Notebooks, GitHub repositories, and cloud computing resources to provide a hands-on, interactive learning experience. The module also includes quizzes and practical exercises to reinforce key concepts, such as selecting search molecules for molecular replacement, using Coot to build protein structures, and refining models with Phenix. By leveraging the NIGMS Sandbox’s cloud-based infrastructure, this module enables researchers and students from diverse institutions to gain practical experience with protein crystallography software and concepts, thereby accelerating their ability to apply these techniques in their own research.

Paper-based clinical protocols and treatments have been historically used by combat medics to treat injuries and optimize survival. There is a need to replace these historical methods with digital processes to modernize the battlespace. With advances in artificial intelligence/machine learning (AI/ML), Chatbot interactions hold the potential to provide critical capabilities for military providers in Multi-Domain Operations (MDO), performing when connectivity is diminished or denied. Chatbots are autonomous, can also refine point of injury treatment, protocols-based patient care procedures, medical history, and current data. The Telemedicine and Advanced Technology Research Center (TATRC) and Madigan Army Medical Center (MAMC) have partnered to explore the feasibility of prototype clinical decision support system (CDSS) Chatbot solutions that feature ML and AI to aid combat medics in clinical protocols and treatment. The current analysis has 3 phases: Phase 1: Available commercial off the shelf (COTS) options were evaluated to determine which COTS technologies were capable of functioning under Delayed/Disconnected, Intermittently Connected, Low-Bandwidth (DIL) MDO conditions. Phase 2: CDSS Chatbot algorithms were trained by the Algorithm Directed 2020 US Army MEDCOM Algorithm-Directed Troop Medical Care (ADTMC) protocols and validated against historical anonymized patient data from previous projects. Phase 3: Chatbot prototype will be integrated with hands-free headset technologies that will be interconnected with the hardware and software solutions acquired in Phase 1. The final prototype will be tested in DIL conditions. Based on the needs assessment conducted in Phase 1, the solutions that offered portable, rugged, and secure devices in DIL/MDO conditions were the Amazon Web Services (AWS) Development Kit (software), AWS Snowball (hardware), Amazon Echo 10 (hands free), and Microsoft HoloLens (hands free) technologies. At the time of this abstract, prototype hardware and software integration into the hands free input devices, Echo 10 and HoloLens, are ongoing. This effort includes development, systematic assessment, and leveraging existing CDSS clinical algorithms into Chatbot enhanced CDSS prototypes that specifically focus on utilizing hands free inputs to provide appropriate medical guidance for casualty treatment.

Image recognition has evolved rapidly from rule-based systems to deep learning models, driven by the exponential growth in visual data and computing power. Traditional on-premise solutions struggle to meet the demands of large-scale, realtime image processing due to limitations in scalability, cost, and operational efficiency. This research addresses the challenge of building a scalable and cost-effective AI image recognition system by leveraging cloud infrastructure. The proposed method integrates deep learning-based image classification with Amazon Web Services (AWS), utilizing EC2 for computation, SQS for asynchronous task handling, and S3 for persistent storage within a modular, auto-scalable architecture. The system demonstrates high throughput, elastic resource management, and reliable classification accuracy across dynamic workloads. Results confirm enhanced performance, cost efficiency, and fault tolerance, making it a viable solution for diverse industries such as healthcare, security, and smart surveillance.

Enterprises now expect the same data platform to serve business-intelligence SQL, relationship analytics and large-language-model–driven semantic search. The practical response is a poly-store that combines a Lakehouse core with graph and vector indexes. Although performance benefits are well documented, quantitative evidence of operational footprint – energy demand, carbon emissions and total cost of ownership (TCO) – is scarce. This paper presents a thirty-day, twelve-hours-per-day benchmark that compares an NVMe-backed ClickHouse cluster with a three-layer prototype (Delta Lake + Neo4j + Milvus) deployed on Microsoft Azure and Amazon Web Services. The workload blends 40 % TPC-DS OLAP queries, 30 % LDBC graph traversals, 20 % ANN-Bench vector searches and a 10 % change-data-capture (CDC) ingest stream. For every 100 000 successful queries were recorded watt-hours via the providers’ Energy/Emissions APIs, dollars at April-2025 list prices and a sustainability-adjusted TCO (TCO-S) that monetises CO₂-equivalent emissions at 80 $ t⁻¹. Under steady load, the poly-store burns around 34 % less electricity and lowers TCO-S by around 27 % thanks to serverless compute de-allocation, specialised query engines and 2.7× columnar compression. A 30-minute CDC surge that quadruples ingest rate doubles both metrics unless tiered SSD caching and simple back-pressure are activated; these mitigations cap the spike at +38 % energy and +31 % cost. Migrating only the object-storage bucket from a high-carbon (around 230 g CO₂e kWh⁻¹) to a low-carbon (around 25 g) region trims TCO-S by a further 11 % without breaching a 100 ms latency budget. The contribution is threefold: the first cloud-native dataset that unites relational, graph and vector modalities with energy metrics, the one-number TCO-S indicator that fuses financial and ESG perspectives and a reproducible experimental setup demonstrating consistent results with minimal variance (≤ 5 % variance). Findings recommend Lakehouse poly-stores for everyday analytics, advise SSD caching for bursty ETL and highlight geography as a low-hanging optimization lever for carbon-aware data platforms.