Narasaraopeta Engineering College is an engineering college for higher education in India, located in Narasaraopeta, Guntur District, Andhra Pradesh, India. Established in 1998, NEC is an Autonomous Institution with permanent affiliation to JNTUK, Kakinada and is promoted and backed by Gayatri Educational Development Society (GEDS).The institution is approved by the All India Council for Technical Education, New Delhi, and has been accredited by the National Board of Accreditation and National Assessment and Accreditation Council with ‘A’ Grade.
The rapid integration of large language models (LLMs) into enterprise systems has introduced new concerns around the exposure of personally identifiable information (PII) during multi-agent interactions. Existing frameworks like CAPRI attempt to mitigate such risks through contextual pseudonymization; however, these methods lack formal privacy guarantees and remain susceptible to inference attacks, particularly in structured reasoning workflows. To address this limitation, we propose CAPRI-DP, an enhanced framework that extends CAPRI with differentially private noise injection applied to sensitive fields within structured entity representations. The approach preserves semantic integrity by applying Laplace perturbations selectively, allowing external LLM agents to operate on masked data without sacrificing task accuracy. Our experiments, conducted across finance and healthcare use cases using ToolEmu-inspired tasks, reveal that CAPRI-DP improves upon fully pseudonymized setups by achieving a 52% success rate—ten percentage points higher—while maintaining fewer conversational turns. Although slightly less performant than entity-only configurations, the proposed method offers a tangible balance between privacy assurance and reasoning effectiveness. These findings suggest that CAPRI-DP can serve as a viable model for privacy-conscious, real-time LLM-agent deployments in sensitive operational domains.
The exponential growth of e-commerce necessitates personalized recommendation systems, especially in fashion domains, where visual attributes heavily influence consumer preferences. This study introduces an ensemble learning model combining five deep learning architectures MobileNet, DenseNet121, Xception, VGG16, and VGG19 to enhance the performance of fashion product recommendation systems. While individual CNN models demonstrated strong classification capabilities, the ensemble achieved superior results with an accuracy of 98.34
Through the development of Internet services, cloud computing, Software-Defined Networking (SDN), Internet of Things (IoT) and fifth generation (5G) communication systems, the number of continually generated network traffic streams is unprecedented. These types of dynamic environments cannot be effectively analyzed by traditional batch learning algorithms, that are inefficient to retrain and cannot effectively deal with concept drift and changing traffic patterns. Online machine learning has become a promising paradigm which continually refreshes the model used for prediction with new observations, which allows realtime traffic classification, anomaly detection, congestion prediction and intelligent network management. This paper is a comprehensive study that compares the most common algorithms used in machine learning which are suitable for network traffic stream analysis in online mode. Comparative framework includes incremental learning capability, computational complexity, memory efficiency, adaptability to concept drift, scalability, classification accuracy, response latency, and suitable for high-speed network environment. In addition, the role of streaming data properties such as feature evolution, class imbalance and resource limitations on the performance of algorithms are discussed. After a comprehensive literature review, this paper presents the advantages and disadvantages, and the practical deployment challenges of widely adopted online learning techniques, and introduces a systematic evaluation framework which could help researchers and network administrators choose the right techniques to implement intelligent, adaptive and scalable network traffic analytics.
DNA computing has been gaining momentum as an alternative paradigm to silicon-based logic, offering unmatched parallelism, low energy consumption, and inherent biocompatibility. However, most DNA-based machine learning models to date have been confined to linear function approximation, limiting their adaptability to real-world decision problems. In this paper, we extend a DNA hybridization-based learning circuit with nonlinear threshold activation implemented via toehold-mediated strand displacement gates, computing a ReLU-like function without enzymes or electronics. To overcome the inability of a single-rail circuit to produce negative outputs, we further introduce a dual-rail encoding scheme in which two complementary strand channels jointly represent signed concentrations. Both variants are validated through mass-action kinetic simulations (Python/NumPy) on an expanded, mixed-sign synthetic dataset (300 training and 60 test samples, four input species), and are benchmarked against an idealized chemical reaction network (CRN) and against representative DNA-based logic and learning systems from the literature. The single-rail circuit's average relative error (ARE) rises to $\approx 95 \%$ under mixed-sign targets because of threshold clipping, whereas the dual-rail circuit converges to $\approx 48 \%$ ARE, closely tracking the ideal CRN (48%), and reduces test-set MAE from 2.11 to 1.14 (RMSE: 2.75 to 1.37). We additionally provide a computational-complexity analysis of strand-species count and reaction latency, and outline a concrete pathway toward wet-lab experimental validation. These findings represent a step forward toward fully autonomous, signed-output DNA-based learning circuits, with direct implications for biosensing, in vivo diagnostics, and programmable therapeutics.
Universities now use Large Language Models (LLMs) to transform their processes for managing student information. The paper introduces an upgraded chatbot system for Narasaraopeta Engineering College (NEC) which extends previous on-premise LLM chatbot research by providing four new functions. The system uses (1) Retrieval-Augmented Generation (RAG) to create citation-based responses through LlamaIndex and ChromaDB, (2) Context Memory which maintains conversation flow during multiple dialogue exchanges, (3) Voice Input through OpenAI Whisper Speech-to-Text (STT) technology, and (4) Multilingual Support which covers English and these seven languages: Hindi, Telugu, Tamil, Kannada, and Malayalam through IndicNLP. The system tested 60 benchmark questions across four academic categories which included regulations and examination policies and fee structures and multilingual queries and achieved 96.7% overall accuracy with sub-second text response times and 1.0–1.4 second voice response times. The system operates entirely on-premise through Docker which safeguards institutional data privacy while eliminating the need for recurring cloud API expenses. The upcoming development will create Emotion-Aware AI, FAQ Auto-Learning, Student Portal Integration, and a Mobile Application.
