Garden City University - School of Computational Sciences and IT, Department of Computer Science
Narra Jishnu Priya Reddy, Sunitha A S
The rapid proliferation of Artificial Intelligence (AI)-powered chatbots is fundamentally reshaping the modern workplace. This paper undertakes a comprehensive investigation into how AI chatbots are transforming workplace productivity, altering employment structures, and redefining human–machine collaboration across diverse industry sectors. Drawing on a mixed-methods design incorporating quantitative survey data (n = 320) and qualitative case analyses, the study explores chatbot impacts on task automation, communication efficiency, decision support, and employee wellbeing. Findings indicate that AI chatbot integration yields measurable productivity gains—averaging 27.4% reduction in repetitive task time—while simultaneously introducing complex workforce disruption patterns affecting clerical and routine customer-service roles. A theoretical model of Human–AI Collaborative Competency (HACC) is proposed, synthesising technology acceptance theory, sociotechnical systems theory, and competency-based HR frameworks. The paper concludes that the future of work necessitates not merely technological readiness but also robust ethical governance, continuous learning ecosystems, and deliberate human-centred design philosophies.
Sarvesh Santosh Gupta, Sandhya N M
Cyber threats have changed a lot in recent years, and the old methods of detecting them just aren’t cutting it anymore. This paper examines how deep learning has been put to work across several security problems — detecting intrusions in network traffic, identifying malware before it spreads, catching phishing attempts, and flagging unusual patterns in logs. But we don’t just cover what’s working. We also didn’t shy away from the problems. Detection models that score beautifully on paper but choke on live traffic. Adversarial inputs crafted specifically to slip past classifiers. Training data that’s either mislabelled, outdated, or just plain scarce. And then there’s the explainability mess — outputs that tell you nothing useful about why a flag was raised. We stopped the search at early 2025. Where the numbers looked too clean, we said so.
Philip John Pereira
Cloud computing is really important for supporting digital applications in many different industries. These days organizations are using cloud infrastructures more and more to process a lot of data and host services that are spread out. So, it has become very important to manage computing resources in a way. The old ways of allocating resources usually rely on fixed rules or mechanisms that use predefined limits, which often do not work well with changing workloads and unpredictable user demand. Because of this cloud environments can have problems like using resources spending more money on operations and potentially having poor performance. Recently there have been advancements in artificial intelligence and machine learning that can help improve how resources are managed in cloud computing systems. By looking at patterns of workload and how the system is performing smart algorithms can predict what resources will be needed in the future and change how resources are allocated in real time. This way of predicting what will be needed helps cloud platforms keep performing while using infrastructure in the best way possible. This paper is proposing a framework that uses artificial intelligence to allocate resources dynamically in cloud computing environments. The framework we are proposing combines real-time monitoring of the system using machine learning to predict workloads and automatic mechanisms to scale resources up or down to make the system more efficient and able to handle more. The architecture of the framework is designed to make decisions, about cloud resource management in a way that adapts to changing needs with the goal of reducing costs. Also, the study talks about how to measure the performance of the proposed approach and what improvements it could bring.
Martin Beishamayum
The rapid growth of smart environments, enabled by the integration of the Internet of Things (IoT), artificial intelligence, and distributed computing, has created a demand for efficient and scalable real-time decision-making systems. Traditional cloud-centric architectures often face limitations such as latency, bandwidth dependency, and delayed responses. This paper proposes an intelligent hybrid framework that combines machine learning techniques with edge computing to support real-time decision-making in dynamic environments including smart cities, healthcare systems, and intelligent transportation networks. Edge devices perform local data preprocessing, anomaly detection, and temporary decision execution, while cloud servers execute deeper predictive analytics and model retraining. The framework employs supervised learning algorithms and lightweight neural networks to optimize computational efficiency and prediction accuracy. Experimental evaluation using real-time sensor datasets demonstrates superior performance in response time, accuracy, and resource utilization when compared with conventional cloud-based systems. The study further highlights cross-domain applications in infrastructure management, healthcare diagnostics, and environmental monitoring. Results confirm that integrating edge intelligence with machine learning significantly improves system reliability, adaptability, and scalability. This work offers a cost-effective and robust architecture aligned with next-generation smart technologies.
Apoorva P
This paper presents a machine learning model using CNN namely ResNet50 for the identification and diagnosis of Soft Tissue Tumours (STTs), specifically skin tumors. The diagnosis of STTs is complicated by their complexity and heterogeneous characteristics, causing potential misclassifications and further emphasizing the need for systems that can provide accurate and timely diagnosis with efficiency. The proposed methodology utilizes high-level data pre-processing and feature extraction to increase classification accuracy and decrease classification error. Specific attention is placed on tumor delineation, as it is critical for treatment planning. The focus of the proposed system is to establish a high level of accuracy in differentiating Melanoma from normal skin tissue, ultimately improving diagnostic accuracy, therapy, and patient outcomes.
Suman Kaman, Saurabh Upadhyay, Gibeon Mushahary
Artificial intelligence has turned prediction into something almost magical—but let’s not get ahead of ourselves. Sure, we’ve moved on from simple rules to deep learning systems that plow through insane amounts of data, trying to predict everything from market moves to game outcomes. But at the end of the day, there’s a simple limit: AI can’t reach into the future. It’s guessing, not seeing. Every forecast is based only on what’s already happened or what’s going on right now. This paper digs into why that is—why AI, fancy as it gets, can only play at being a fortune teller. Concepts like aleatoric and epistemic uncertainty, the one-way flow of time, and the stubborn facts of causality all keep the future locked away. AI can simulate or guess with high probability, but there’s a kind of hard wall—an Ontological Temporal Barrier—it just can’t cross. Real-world cases show how easy it is to mix up strong guesses with genuine foresight, but once you look closer, the future is still a closed book until it shows up as today’s news.
Punith R
Artificial Intelligence (AI) has emerged as a transformative technology in healthcare, particularly in the early detection, diagnosis, and management of diseases. Leveraging advanced machine learning, deep learning, explainable AI (XAI), ensemble learning, and edge computing techniques, AI enables the analysis of vast and complex biomedical datasets. This paper presents a comprehensive review of recent developments and applications of AI across multiple disease domains, including cardiovascular diseases, chronic kidney disease, retinal disorders, COVID-19, and plant diseases used as proxy models for XAI research. Key methodologies examined include convolutional neural networks (CNNs), transfer learning, generative adversarial networks (GANs), federated learning, Long Short-Term Memory (LSTM) networks, and ensemble models (Random Forest, XGBoost, CatBoost, LightGBM). The paper further examines emerging trends such as the integration of AI with the Internet of Medical Things (IoMT), blockchain for secure data sharing, and augmented reality for enhanced clinical decision support. Research gaps including limited explainability, dataset scarcity, clinical workflow integration, and multi-modal data fusion challenges are identified and discussed. This review provides a holistic perspective on the capabilities, challenges, and future directions of AI-driven disease detection, underscoring its potential to revolutionize healthcare delivery.
T Mohammad Shadik
Career decision-making is one of the most critical yet challenging aspects of a student’s life. Traditional counselling methods are often generic, outdated, and inaccessible, resulting in confusion and skill-job mismatches. This paper presents the Career Path Explorer using AI, an intelligent platform that leverages artificial intelligence to provide personalized, data-driven career recommendations. The system integrates skill and personality assessments, academic performance, and real-time labour market trends to generate adaptive career pathways. Prototypes were developed using design thinking methodology, tested with students and educators, and refined iteratively. Pilot validation indicated strong acceptance, with 85% of students preferring AI-driven recommendations over conventional counselling. The platform demonstrates significant potential for improving career decision-making, reducing mismatches, and contributing to global educational initiatives such as NEP 2020 and UN SDG 4.
Preethu S, Kushal B S
Skin cancer is one of the most common kinds of cancer in the world, and the early detection of this cancer plays a major part in the improvement of patient survival rates. Automated skin lesion classification with the help of deep learning has proved to be a promising decision support tool, which may help a dermatologist to identify a potentially malignant lesion from a dermoscopic image. DermatoNet, a deep learning framework, has been developed based on transfer learning, which classifies images of skin lesions based on a dataset. The images used in the dataset are of seven pigmented skin lesions: Melanocytic Nevus (Benign Mole), Melanoma (Malignant Skin Cancer), Benign Keratosis, Basal Cell Carcinoma, Actinic Keratoses/Intraepithelial Carcinoma, Vascular Lesions, and Dermatofibroma. Most of the dermatology datasets contain a limited number of samples, which results in an imbalanced dataset, which in turn reduces the accuracy of classification. This may be achieved with the help of transfer learning, which enables the effective training of a model even with a limited number of samples in the medical images.
Uday Rajesh Thanki, Sandhya N M
Traditional paper business cards are not very good because the information on them does not change and you have to pay to print them again and again. Also, it is not easy for people to get the information from them. This research paper is about a digital business card platform called Phygital. We made Phygital using the MERN stack. This platform lets users make their digital business cards and share them with others using QR codes and public profile links. We used React to make the frontend of the platform, Node.js and Express for the backend and MongoDB to manage the database. We also used Cloudinary to store media and Render to deploy the platform. The Phygital platform is very secure. It is easy to use on different devices. It also tracks how people use it. Helps people network with each other in a better way. The results show that Phygital makes it easier for people to get the information they need reduces the need for paper and helps people connect with each other efficiently. The Phygital digital business card platform is a solution, for people who want to network with others.
Sibi Jaidan B S
The internet and social media have pretty much flipped the script on how we get and share information. Now, it’s easier than ever to find what you’re looking for, but at the same time, it’s just as easy for fake news and misinformation to spread like wildfire. Fake news isn’t just sloppy reporting, it’s when someone actually creates or shares stuff that’s made up or meant to mislead, all while making it look like real news. This isn’t just annoying; it can mess with politics, shake up economies, and even put people’s health at risk. Traditional fact-checking eats up a lot of time because experts have to go through everything by hand. With the flood of digital info out there, that just doesn’t cut it anymore. So, researchers in computer science and data science have started focusing on ways to automate fake news detection. That’s where Natural Language Processing, or NLP, comes in. It gives machines the tools to actually make sense of human language, which makes it a game changer for spotting fake news. This paper looks at how Natural Language Processing helps spot fake news online. It digs into the text of news articles, using tools like tokenization, stop word removal, stemming, and lemmatization to clean things up. Then, it pulls out features with methods like TF-IDF and word embeddings, turning the words into numbers that machine learning models can actually work with. The study tests out different algorithms, Logistic Regression, Naïve Bayes, Support Vector Machines, and even deep learning models like LSTMs, to see how well they can tell real news from fake. The experiments show that NLP models do a great job telling fake news apart from the real stuff. When you bring in advanced deep learning models like those fancy transformer architectures, they push the accuracy even higher, thanks to their knack for picking up on context in the text. All of this really drives home how crucial automated fake news detection is if we want to fight misinformation and keep digital information trustworthy.
Antony Abhishek A
Responsive Web Design (RWD) is essential for delivering a consistent user experience across devices with different screen sizes. This study evaluates the impact of RWD on usability, accessibility, readability, and user engagement by comparing responsive and fixed-layout websites. The results show that responsive websites improve task completion time, navigation, and overall user satisfaction while reducing cognitive load and navigation errors. Features such as flexible layouts, media queries, adaptive grids, and mobile-first design enhance cross-platform compatibility and accessibility, including support for assistive technologies. The findings also indicate that RWD improves customer retention, search engine visibility, and digital credibility. Overall, the study concludes that Responsive Web Design is a key practice for developing efficient, accessible, and user-centered websites in modern digital environments.
Rabindra Rai
Software quality assurance plays a vital role in ensuring the reliability, maintainability, and performance of software systems. This paper presents a Machine Learning-based approach for software defect prediction using classification algorithms such as Decision Tree, Random Forest, Support Vector Machine, and Naïve Bayes. The proposed model analyzes software metrics including code complexity, lines of code, coupling, cohesion, and change history to predict potential defects in software modules. Data preprocessing techniques such as normalization, feature selection, and handling missing values are applied to improve prediction accuracy. The performance of the proposed system is evaluated using metrics such as accuracy, precision, recall, and F1-score. Experimental results indicate that Random Forest and Decision Tree algorithms provide higher prediction accuracy compared to other models. The study demonstrates that Machine Learning techniques can significantly support software testers and developers in prioritizing testing efforts, reducing maintenance cost, and improving overall software quality. This research highlights the importance of intelligent defect prediction systems in modern software development environments.
Greeshma BL, Prof. Pallabee Padhi
Machine learning is rapidly transforming the cybersecurity landscape by enabling intelligent systems to detect and prevent online threats. One of the most significant threats today is phishing, where attackers create fraudulent websites to steal sensitive information such as passwords and financial data. Traditional rule-based detection systems often fail to identify new and sophisticated phishing attacks. This paper aims to provide a comprehensive overview of a Machine Learning-Based Phishing Detection System for enhancing cybersecurity. It highlights recent advancements in machine learning techniques for phishing detection from 2022–2025, including improved accuracy, real-time detection, and adaptive learning models. Furthermore, the challenges such as dataset imbalance, feature selection, evolving attack patterns, and model generalization are discussed, along with possible solutions to overcome these limitations.
Pramodh Kumar K S
Security plays an important role in human life and endeavours. Securing information and disseminating it are critical challenges in the present day. Victims and witnesses of cybercrime often hesitate to report incidents due to concerns over privacy, complexity, and fear of retaliation. Traditional reporting mechanisms require manual data entry, creating accessibility barriers and delaying response times. To address these challenges, this paper introduces an AI-driven voice-based cybercrime reporting system that allows victims and witnesses to anonymously submit complaints through audio recordings. Leveraging speech recognition transformers, recent language models, and encryption, the system processes real-time multilingual voice inputs, extracts meaningful content, and classifies reports with high precision using a hybrid voting mechanism. The study recommends that governments, organisations, and individuals should emphasise moral development, regular training of employees, regular software generation updates, use of strong passwords, data backups, strong cybersecurity policies, antivirus software and security surveillance (CCTV) in offices to safeguard employees and properties from being hacked and vandalised. Experimental evaluations on synthetically generated and human-validated datasets confirm the system’s ability to accurately transcribe, classify, and securely process audio complaints while preserving user anonymity. This work improves cybercrime reporting by making it more accessible, efficient, and secure, fostering greater participation from victims and witnesses.
Niveditha Naik
Personalization plays a key role in improving user satisfaction in digital platforms, especially within the tourism and food service sectors. Traditional recommendation systems often operate within a single domain, limiting their ability to provide comprehensive suggestions that match diverse user preferences. This paper proposes a personalized recommendation system for tourism and food services using a hybrid filtering approach that integrates collaborative filtering and content-based filtering techniques. The model analyzes user behavior, historical preferences, item attributes, and contextual factors such as location and time to generate accurate and relevant recommendations. The hybrid filtering method overcomes the limitations of individual techniques by combining similarity-based user patterns with feature-driven item analysis. Experimental results demonstrate enhanced precision and relevance in recommendations when compared with standalone models. The proposed system can be effectively implemented in travel platforms, food delivery applications, and tourism service aggregators to deliver a rich, personalized user experience.
Priyanshu Singh, Dr S.Narmada
Retrieval-Augmented Generation (RAG) is one of the most significant architectural advances in modern Natural Language Processing. It solves a fundamental problem with Large Language Models (LLMs): their knowledge becomes outdated and they sometimes generate incorrect information — a problem called hallucination. RAG fixes this by connecting the generative model to a searchable external knowledge base, so responses are grounded in real, retrieved evidence. This paper provides a comprehensive analysis of RAG system architecture, retrieval strategies, evaluation frameworks, and deployment across real-world applications. We examine five RAG variants — Naive RAG, Advanced RAG (DPR), FiD, Self-RAG, and HyDE+Hybrid — and compare them on standard benchmarks. Results show that the best RAG pipeline reduces hallucination by up to 76.7% compared to a standard LLM and achieves a RAGAS composite score of 0.85. Hybrid retrieval methods consistently outperform simpler approaches. Open challenges and future research directions are also discussed.
Sujan Gopi S, Prof. Sahana Edwin
Vehicle Vault is an integrated web-based platform designed to digitize and streamline the complete lifecycle of vehicle ownership, maintenance, and resale. The system consolidates vehicle registration records, service history logs, insurance tracking, and a peer-to-peer marketplace into a single secure application. By leveraging modern web technologies including React.js, Node.js, and a cloud-hosted MongoDB database, Vehicle Vault provides real-time access to vehicle data for owners, mechanics, dealers, and regulatory authorities. This paper presents the system architecture, key modules, implementation methodology, and experimental results demonstrating improved data accessibility, reduced administrative overhead, and enhanced transparency in used-vehicle transactions.