Building an Intelligent Research Assistant – RAG Isn't All You Need

IntroductionMaintaining patient safety during transition from prehospital to emergency department (ED) care depends on effective handoff communication between providers. We sought to determine emergency physicians’ (EP) knowledge of the care provided by paramedics in terms of both procedures and medications, and whether the use of a verbal report improved physician accuracy.MethodsWe conducted a 2-phase observational survey of a convenience sample of EPs in an urban, academic ED. In this large ED paramedics have no direct contact with physicians for non-critical patients, giving their report instead to the triage nurse. In Phase 1, paramedics gave verbal report to the triage nurse only. In Phase 2, a research assistant (RA) stationed in triage listened to this report and then repeated it back verbatim to the EPs caring for the patient. The RA then queried the EPs 90 minutes later regarding their patients’ prehospital procedures and medications. We compared the accuracy of these 2 reporting methods.ResultsThere were 163 surveys completed in Phase 1 and 116 in Phase 2. The oral report had no effect on EP awareness that the patient had been brought in by ambulance (86% in Phase 1 and 85% in Phase 2.) The oral report did improve EP awareness of prehospital procedures, from 16% in Phase 1 to 45% in Phase 2, OR=4.28 (2.5–7.5). EPs were able to correctly identify all oral medications in 18% of Phase 1 cases and 47% of Phase 2 cases, and all IV medications in 42% of Phase 1 cases and 50% of Phase 2 cases. The verbal report led to a mild improvement in physician awareness of oral medications given, OR=4.0 (1.09–14.5), and no improvement in physician awareness of IV medications given, OR=1.33 (0.15–11.35). Using a composite score of procedures plus oral plus IV medications, physicians had all three categories correct in 15% of Phase 1 and 39% of Phase 2 cases (p<0.0001).ConclusionEPs in our ED were unaware of many prehospital procedures and medications regardless of the method used to provide this information. The addition of a verbal hand-off report resulted in a modest improvement in overall accuracy.

This scientific article is devoted to an in-depth study of the role, potential, and prospects for the application of artificial intelligence in the development of modern digital marketing. The paper substantiates that digital marketing is undergoing a fundamental transformation, shifting from traditional mass advertising models to data-driven, customer-centric strategies based on the processing and analysis of large volumes of structured and unstructured data. Artificial intelligence is considered a key technological driver of these changes. The article examines modern artificial intelligence tools used in digital marketing and analyzes their impact on the automation of marketing processes, the improvement of analytical accuracy, and the formation of personalized consumer experiences. Particular attention is paid to the integration of machine learning algorithms, natural language processing technologies, and generative artificial intelligence into enterprise marketing activities. The study highlights the role of AI in predictive analytics, customer behavior forecasting, dynamic content generation, recommendation systems, chatbots, and virtual assistants, which significantly enhance the efficiency of interaction with target audiences. The advantages of using artificial intelligence in the processes of hyper-personalization of marketing communications, optimization of advertising campaigns, and real-time decision-making are emphasized. At the same time, the paper critically analyzes the key challenges and limitations associated with the implementation of AI in digital marketing. These include issues of data privacy and cybersecurity, ethical concerns, algorithmic bias, transparency of automated decision-making, and the lack of a unified and comprehensive regulatory framework governing the use of artificial intelligence technologies. Based on the analysis of current scientific research and practical business cases, the article identifies the main trends in the further development of digital marketing, including the expansion of hyper-personalization, the growing importance of voice and visual search technologies, and the increasing share of content generated by artificial intelligence. The study concludes that the effective, ethical, and responsible implementation of artificial intelligence tools is a critical prerequisite for enhancing enterprise competitiveness and ensuring sustainable development in the digital economy.

This study proposes a modified and adaptive Balanced Scorecard (BSC) framework tailored to the strategic context of art museums as its primary contribution. The adapted BSC categorizes museum strategies into four dimensions—Social Values and Cultural Impact (SVCI), Visitor Experience and Tourism Development (VED), Exhibition and Operational Efficiency (EOE), and Art/Cultural Education and Sustainable Growth (ACES)—providing a systematic lens for interpreting and quantifying qualitative managerial priorities. These strategic perspectives were extracted from unstructured textual data sourced from museum websites and visitor-generated online content through natural language processing and text mining techniques. Building on this strategic classification, the study develops deep learning–based predictive models that integrate both structured numerical data and text-derived qualitative variables to estimate visitor numbers. Eight deep learning architectures, including Transformer, Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Generative Adversarial Network (GAN), were evaluated. The results indicate that VED exerts the most substantial influence on visitor attendance, particularly through interactive exhibitions, personalized experiences, and tourism collaborations, while other strategic dimensions provide marginal improvements in predictive accuracy. The Transformer model achieved the best predictive performance, reflecting its superior ability to capture complex relationships between structured and unstructured data. By combining an adapted BSC framework with advanced predictive modeling, this study offers both a novel theoretical approach for categorizing museum strategies and a practical tool for enhancing decision-making in visitor engagement, exhibition planning, and resource allocation.

Chapter 7 - Data Structures and the Anatomy of a File

Objective - To investigate the types of questions students ask and the language they use in virtual reference. It is hoped that this examination will provide understanding of students’ needs and thus improve/enhance library services. Methods - Over 600 virtual reference transcripts were reviewed, analysed and categorised. This work was focused on three levels of analysis: broad categories based on the general type of question being asked, subcategories based on the specific question and the language that students used to ask their questions. Results - Students are primarily using the library’s virtual reference service for higher-level research assistance rather than using the tool to obtain quick answers to simple questions. The two most common types of questions involved staff providing detailed information or instruction on a topic. More specifically, the most frequently occurring type of question was related to finding journal articles on a given topic. Our analysis of the words students use to ask their questions confirmed that students and librarians often do not speak the same language. Conclusion - The results of our analysis of students’ needs and language can help us understand our users. This study demonstrated that our library can enhance services in five areas: online services, collections, relationships, staff skills, and the library as place.

Intelligent manufacturing technologies, including databases, mathematical modeling, and information systems have played a significant role in process control, production management, and supply chain management in traditional Chinese medicine(TCM) industry. However, their ability to process and utilize unstructured data, such as research and development reports, batch production records, quality inspection records, and supplier documents, is relatively weak. For text, images, language, and other unstructured data, generative artificial intelligence(AI) technology has shown strong potential for development in extracting information, extracting knowledge, semantic retrieval, and content generation. Generative AI is expected to provide a feasible set of tools for the utilization of unstructured data resources in the TCM industry. Based on years of research and industrial application experience in TCM intelligent manufacturing technology, this study reviewed the current situation of intelligent manufacturing in TCM and the utilization of unstructured data, analyzed the application value of generative AI in the TCM manufacturing process and supply chain, summarized four typical application scenarios, including intelligent pharmaceutical knowledge base/knowledge graph, intelligent on-the-job trai-ning, intelligent production quality control, and intelligent supply chain. Furthermore, this study also explained the data collection and processing, business process design, application potential, and value of each scenario based on industry demands. Finally, based on the integration of generative AI and TCM industrial models, the study proposed a preliminary concept of a smart industrial brain for TCM, aiming to provide a reference for the application of AI technology in the field of TCM manufacturing.

The number of users of health systems on the Internet has increased at a rapid pace. The systems are browsed for health information, and online interactions take place with medical experts as well as with people who have similar health problems. This paper presents a study of the usage of two different types of health systems: Ask the Expert system and online communities for people with overweight. The study focused on the different types of questions as well as information, advice and recommendations posted in the systems. Starting point of the study was the questions and answers presented in the Ask the Expert system. The results show how the systems serve as complements, and suggestions on how to enable the users to easily cross-use the systems are presented. Such development could help the users to get a fuller picture and better understanding of different health issues.

An increase of production activity effi ciency of an enterprise can be reached by effi ciency increase of the management decisions making process. Modern highly intellectual management systems are based on the principles of fuzzy logic and allow not only to process the large volumes of versatile and unstructured information, but also to make decisions quickly in conditions of invariance and insuffi ciency of data. Such systems usage and adaptation to real working conditions are main ways to increase the production effi ciency. The article describes the possibility of operational management system design. Such system is based on symbiosis between multi-level multi-agent systems and dynamic expert systems and is expected to be used for processes of section mill stand preparation and transfer, making and adjustment of the rolling plan under the conditions of interaction of all agents of a technological circuits.

LAMB: An open-source software framework to create artificial intelligence assistants deployed and integrated into learning management systems

Task modelling helps to ensure the usability of interactive systems by producing models that are useful for a variety of reasons, and by forcing designers to view the system from the users’ point of view. Unfortunately, it is a difficult process, so most designers either don’t do it at all, or they model only certain aspects of the system, potentially using different formalisms that are hard to consolidate. As a result, acomplete and coherent model of all the levels of a system is seldom if ever built. To address this problem, this paper proposes a flexible methodology for the construction of task models and an environment that supports this methodology. The methodology, which is based on industrial experience, identifies a sequence of levels for which models should be built, and the environment provides tools for building them, either from scratch or from existing sources.

Mark Ambler, BSN, MBA, RN, CCRN, CNML

A growing body of evidence over the past twenty years suggests that electronic prescribing (ePrescribing) can bring a host of benefits to the health sector, and some policy makers now view ePrescribing as a desirable national service that can improve the safety, quality and efficiency of healthcare. Consequently, national ePrescribing and electronic health record (EHR) projects have been launched in many countries in recent years, but success is elusive as ePrescribing in primary care is still unavailable in over 90% of countries worldwide. Progress has been limited to a small number of countries, with a significant cluster in Europe. Why is the adoption of ePrescribing and EHRs slow and uneven? What are the factors that influence the development and adoption of ePrescribing? What are the building blocks for a national ePrescribing service? What are the lessons from the leading countries? What are the policy lessons for countries where national ePrescribing services are not yet developed? This research addresses these questions. Qualitative methods were used, including a comparative analysis of case studies in thirty one European countries and interviews with national ePrescribing experts in the leading countries. A cross-domain comparison was undertaken with a leading eGovernment service in Ireland - the Revenue On-Line Service (ROS) – and interviews were held with national Revenue experts. The context, content and process framework from Pettigrew and Whipp’s model of strategic change and Roger’s diffusion of innovations model provide the conceptual frameworks for analysis and comparison. The research found that ePrescribing services in primary care are now available on a national basis in eleven countries in different regions of Europe. A model for national ePrescribing is emerging over time, which includes local EHR integration, interoperability between prescribers and pharmacists through a national ePrescription database, and a trend towards on-line patient access to medication data using the Internet. In many other European countries ePrescribing and in particular the electronic transmission of prescriptions (ETP) is underdeveloped due to a range of organisational, legal, technical or other barriers. In these countries, the absence of a developed national ETP service is the greatest barrier to adoption. The adoption patterns of recent years suggest that a technical-legal ePrescribing divide exists in Europe between the leading and the following countries. The comparative analysis discovered that a strong inner context is the cornerstone of success in the leading countries and ROS. A strong inner context can shape the outer context, develop quality content and create a supportive process for adoption and diffusion. The findings suggest that this can be achieved through a complex process of continuous stakeholder management. Building blocks were also identified in the outer context in the legal and technological areas, where the national health model in operation emerged as the key contextual factor. The quality of the content was found to be a key adoption factor, including the interoperability, security and usability of ePrescribing systems and ETP services. The key critical success factor for the adoption of ePrescribing and ROS was found to be stakeholder management in the adoption process over a long period. Supporting factors included social marketing strategies, incremental implementation methods, quality user support and the judicious management of consent, incentives and mandates. The overall approach used by the leading countries emerges as a balance of central control and direction from the inner context, coupled with widespread permanent stakeholder management across the domain including the outer context and the private sector, leading to a complex process of development and adoption over a long period. This equates to a combination of the traditional top-down and bottom-up approaches, which Coiera has described as middle-out. The findings in this research may contribute to policy in this area, Irish policy in particular.

Context-aware computing is a growing research domain in present circumstances due to technological advancements in the area of sensors technology, big data, artificial intelligence and robotics and automation. It has many applications for making the daily life of human beings sustainable, comfortable, and smooth. Context ware computing also includes ambient intelligence and applications such as occupancy detection, prediction, user recognition etc. Occupancy detection and recognition help in developing intelligent applications which help the energy management, intelligent decision making, that results in cost reduction and fault and failure prevention of services and products in advance. Several studies have been conducted to detect the occupancy with a different set of methodologies and approaches using varying types of data such as environmental parameters, image and video-based attributes, wireless or sensor based parameters, and noise-based parameters. This paper proposes a reliable, more accurate and efficient model based on the statistical analysis of the sensor based data for occupancy detection. Detailed quantification of the relationship of the ambient attributes is presented and the ensemble model is developed based on machine learning technique extreme learning machine to achieve the significant level of improvement in accuracy, efficiency, generalization and reliability. In addition to this, the paper also proposes one online and adaptive model-based online sequential extreme learning machine to perform occupancy detection on real-time data when complete data is not available and learning is done with recent data points coming in the form of streams. Results are compared with existing work in the domain and it is observed that proposed model perform better in terms of efficiency and accuracy over existing literature work.

In the fast-paced and precision-driven environment of well construction operations, efficient decision-making is paramount. This paper introduces an advanced artificial intelligence (AI) model leveraging Large Language Models (LLM) to classify log records and complex WITSML data in real time. By integrating this AI model into drilling control rooms, decision-making efficiency has been enhanced by over 50 times, significantly reducing the time required for search and background analysis from hours to minutes. The implementation of LLMs such as GPT-3 and GPT-4 has revolutionized the process by enabling zero-shot inference, allowing for the accurate classification of complex texts without extensive labeled training datasets. This breakthrough has facilitated the rapid processing and analysis of structured and unstructured data, which is crucial for real-time response to critical events in drilling operations. A collaborative effort between drilling experts and digital developers resulted in a functional proof of concept within two weeks, expanding to a minimum viable product (MVP) in under 90 days. The AI model now categorizes technical texts into 42 distinct categories, surpassing initial goals and demonstrating substantial improvements in both accuracy and processing speed. This paper concludes by highlighting the significant operational efficiencies and data consistency achieved through the AI model. It also discusses potential future enhancements, including the integration of diversified data sources and the development of predictive and prescriptive models, positioning this AI-driven approach as a cornerstone for future advancements in well construction control rooms.

The ability to accurately classify service types from technician comments is a critical challenge for many automotive original equipment manufacturers (OEMs) that lack standardized opcode systems. Existing rule-based methods, while precise, leave numerous observations unlabeled due to variability in textual descriptions. This study proposes a novel approach that combines text mining techniques with a multinomial classification model to automate service type classification. By leveraging structured and unstructured data, this method achieves a significant improvement in classification accuracy, laying the groundwork for enhanced analytics and predictive modeling in automotive service data.