Authors - Vedant Chandore, Niranjan Pardeshi, Sai Sinare, Samruddhi Akude, Sahil Dhawane, Kartik Gawande, Rahul Sadgir, Shravani Nigade, Ajay Talele Abstract - For transportation infrastructure to be safe, effective, and long-lasting, road condition monitoring is essential. Conventional techniques, which depend on human inspections, are frequently ineffective and prone to mistakes. To overcome these constraints, this study suggests a machine learning-based smart road condition monitoring system. Utilizing cameras installed on vehicles, the system gathers pictures and videos of the state of the roads, which are subsequently processed by sophisticated machine learning algorithms. These algorithms categorize surface conditions, identify irregularities in the road, and offer information on repair requirements. Through comprehensive field testing and data analysis, the study shows how effective the system is, showing notable gains in both the efficiency of maintenance procedures and the accuracy of identifying road issues. By concentrating on image and video analysis, this smart monitoring system offers a revolutionary solution for urban infrastructure management, opening the door for safer, more intelligent, and sustainable road maintenance procedures. This strategy not only lowers operating costs but also improves road safety and infrastructure sustainability.
Authors - Sai Himagnya Parisaneni, Vemula Surya Teja, Revanth Guthula, Sushama Rani Dutta Abstract - This study presents an optimized approach for detecting mental disorders by integrating support vector machines (SVM) enhanced through Minimum Bayes Error Rate (MBER) optimization. The proposed framework uses MBER Optimization and refines classification boundaries through SVMs improve decision-making. Unlike conventional deep learning approaches that rely solely on CNN based end-to-end learning, our method uses SVM for classification that minimizes errors, enhancing model robustness and generalization. The experimental evaluation on EEG-based datasets assesses the effectiveness of the hybrid approach in terms of accuracy, computational efficiency, and scalability. The results provide insights into the potential of MBER-optimized SVM models for real-world applications in mental health diagnostics.
Authors - Satish Chikkamath, Shreya Pattanashetti, Pooja V Gadad, Vidya Revanakar, Bhoomika Hosamani Abstract - Emojis serve as an established means for people to express emotions and sentiments while interacting on social media. This paper examines the task of emoji prediction from text by developing accurate classification methods. The model uses a pre-trained and fine-tuned BERT framework on a dataset consisting of text sentences along with their corresponding emojis. This structured data allows the model to capture contextual meaning and emotional nuances, which are crucial for practical applications. Challenges associated with emoji usage are addressed through tokenization techniques in text preprocessing, while performance advances are achieved using stemming and feature extraction. Research conclusions indicate that the BERT-based model outperforms traditional deep learning approaches like LSTM. This study spotlights how NLP and sentiment analysis contribute to emoji prediction and shows its practical applications in social media monitoring, sentiment analysis, and enhancing user experiences.
Authors - Sagar Janokar, Krish Deshpande, Krishna Masane, Shriyash Kothe, Varad Kulat, Krish Chabria, Rushikesh Kuchekar Abstract - This project demonstrates how a machine learning based approach can revolutionize the analysis of unstructured text data in defense intelligence. By automating key processes, the system will enable faster and more accurate identification of threats and patterns, improving decision-making and operational efficiency. This innovative application highlights the transformative role of technology in addressing real-world challenges in intelligence gathering.
Authors - Dhanaselvam J, Dhanalakshmi R, Prashaanth S, Hariprasath S, Harish R Abstract - India, with three-fourths of its population dependent on agriculture, is plagued by severe crop loss due to pest infestation, particularly in staple crops like rice, wheat, maize, and soybeans. This paper proposes an embedded system of real-time pest detection and precise pesticide spraying to enhance productivity. The system employs deep learning with a Residual Neural Network (ResNet) and Quadra-attention, residual, and dense fusion techniques for enhanced pest image classification. High-resolution images of crop leaves are captured, pre-processed, and analyzed for pest detection. Upon detection, the system selects the appropriate pesticide and activates an autonomous robotic sprayer. Driven by an Arduino NANO-based module with an L293D motor driver, the robotic system automatically navigates through fields, ensuring precise pesticide application without waste and infrastructure costs. With IoT integration and 99.80% validating accuracy, this system optimizes pesticide use, enhances crop health, and enhances yield, offering a cost-effective automated pest management system for sustainable agriculture.
Authors - Nikita Bhatt, Nirav Bhatt, Purvi Prajapati Abstract - In today’s data-rich world, we often deal with multiple types of information such as images, text, and audio. Traditional deep learning models usually focus on a single type of data, but real-world applications need systems that can understand and connect across these different formats — a concept known as multi-modal learning. This paper explores cross-modal retrieval, where a user can input one type of data (like an image) and retrieve another (like related text). To make this possible, we map different data types into a common space using deep learning methods like CNN for images and LSTM for text. One of the key challenges in this area is comparing vectors of different lengths, which affects similarity estimation. Most traditional methods use inner product similarity, which is not ideal for vectors with varying magnitudes. To overcome this, we normalize the vectors using cosine similarity, which focuses only on the angle between vectors, not their length. This improves retrieval accuracy by reducing noise caused by vector size differences. We also discuss the benefits of using deep learning to jointly learn features and generate hash codes for faster and more accurate retrieval. Experiments on datasets like Google News show that cosine similarity outperforms Euclidean distance in terms of retrieval performance, especially when combined with models like CBOW.
Authors - Ria Ashish Gawali, Christopher Sachin Chopde, Aryan Gupta, Tashmeet Kaur Jasbeersingh Hora, Rachna Karnavat Abstract - MediaGPT is a Generative AI system that combines natural language and image synthesis to create unified, visually appealing media content. By combining strong language models such as ChatGPT and Phi-3 with image synthesis models such as Stable Diffusion and ControlNet, MediaGPT facilitates intelligent text-image alignment on an interactive canvas. The layout can be easily customized along with semantic coherence and aesthetic balance. Developed for designers, educators, marketers, and content creators, MediaGPT improves the creative process by facilitating effortless multi-modal integration and providing easy-to-use tools for creating high-quality, contextually appropriate content.
Authors - Lokesh Khedekar, Atharva Kassa, Kartavya Sharma,Tejas Kedar, Sarthak Kasar, Kaustubh Kelgandre, Sharad Kasralikar Abstract - Natural Disasters have been a major threat to the living beings, environment and the infrastructure, in mainly areas where they have poor access to early warnings systems. This paper provides AI-based Natural Disaster Response System which helps to evaluate the impact of natural disasters and gives better of the existing systems. The system has historical Geographic Information System (GIS) datasets with real-time data from Internet of Things (IoT) sensors and predictive modeling to check out the natural disaster’s magnitude, area of impact, and resources. The methodology includes data preprocessing, feature extraction, and machine learning model training to achieve effective predictive accuracy. A Convolutional Neural Model (CNN) model was created and tested which further achieved 93% accuracy of predicting the impact of the disaster incident. The system was then compared with other machine learning models, then was proved to be more effective. The suggested method gives efficient, cost-effective and scalable way of utilizing the emergency resources at the maximum.
Thursday August 27, 2026 3:30pm - 5:30pm IST Virtual Room EGOA, India
Authors - Prasanna Lakshmi T, Shankar Lingam. M Abstract - This paper explores the intersection of emerging technologies and ICT policy evolution in India, with a focus on Artificial Intelligence (AI), blockchain, the Internet of Things (IoT), and 5G technologies. As India navigates its digital transformation through initiatives like Digital India, the paper examines how the nation's ICT policy framework has adapted to accommodate these disruptive technologies. Using a theoretical approach based on Technological Innovation Systems (TIS), the study traces the historical development of India's ICT policies, from early telecom regulations to the modern-day focus on digital infrastructure and smart technologies. Challenges such as the digital divide, cybersecurity, and data privacy are also analyzed. By identifying key policy milestones and evaluating India's current efforts in integrating emerging technologies, this paper provides insights into the future direction of ICT policy in India. The findings highlight both opportunities and barriers to sustainable technological advancement and offer policy recommendations to better align ICT governance with global trends.
Authors - Agatsya Yadav, Renta Chintala Bhargavi Abstract - Large Language Models (LLMs) offer powerful capabilities but their significant size and computational requirements hinder deployment on resource-constrained mobile devices.This paper investigates Post-Training Quantization (PTQ) for compressing LLMs for mobile execution. We specifically apply 4-bit PTQ using the BitsAndBytes library via the Hugging Face Transformers framework to Meta’s Llama 3.2 3B model. The quantized model is further converted to the GGUF format using llama.cpp tools for optimized mobile inference. The proposed PTQ workflow achieved a 68.66% reduction in model size through 4-bit posttraining quantization, enabling the Llama 3.2 3B model to run efficiently on a standard Android device. Qualitative validation confirmed the 4- bit quantized model’s ability to perform inference tasks successfully. We demonstrate the feasibility of running the final quantized GGUF model on an Android device using the Termux environment and the Ollama framework. PTQ, particularly down to 4-bit precision combined with mobile-optimized formats like GGUF, presents a viable pathway for deploying capable LLMs directly on mobile devices, balancing model size and functional performance.
Thursday August 27, 2026 3:30pm - 5:30pm IST Virtual Room EGOA, India
Authors - Dwayne Nixon, Shaun Menezes, Ramya Kulkarni, Phiroj Shaikh Abstract - In today’s fast-paced development environment, where efficiency and speed are paramount, manual tasks such as taking screenshots, converting files, rebooting systems, and managing repositories have become increasingly tedious and time-consuming. These routine activities disrupt developer workflow and hinder productivity, consuming valuable time. To address these inefficiencies, this work proposes a comprehensive automation tool that extends beyond handling basic tasks to streamline workflows and optimize productivity. Firstly, this tool centralizes a wide range of operations, including automating code generation, creating detailed reports, and developing websites. By integrating these functionalities, developers can eliminate redundant tasks and focus on high-level problem-solving. Secondly, the automation tool enhances accuracy and consistency across development projects, ensuring higher standards of work and reducing errors associated with manual processes. Furthermore, the tool aligns with evolving technological demands, enabling teams to adapt to increasing project complexities while maintaining efficient workflows. This solution represents a transformative approach to software development, combining automation and centralization to reduce manual workloads and optimize developer productivity. The implementation of such an all-in-one automation platform promises to significantly improve efficiency and foster innovation in the industry.
