Publications

This work presents an advanced automated system designed to streamline the processing of medical prescriptions. The system follows a multi-stage workflow aimed at enhancing efficiency in handling and interpreting prescription data. The process begins when a user uploads a prescription image, which is then subjected to a series of sophisticated image processing techniques, including grayscale conversion, noise reduction, and morphological operations. These steps are crucial for improving the readability of the text within the image, ensuring that the subsequent stages operate on high-quality data. Once the image is processed, an Optical Character Recognition (OCR) module is employed to extract text from the image. The OCR process involves text region detection, segmentation of lines and characters, and the application of algorithms to recognize and transcribe the text accurately. Following OCR, the extracted text is further refined using language modeling techniques, which involve spell checking, error correction, and the recognition of key entities such as medication names and dosages. To provide meaningful insights, a large language model (LLM) is used to generate a concise and coherent summary of the prescription. This summary distills the essential information from the prescription, making it easier for healthcare professionals and patients to understand and manage the prescribed treatments. The final summary is then returned to the user, completing a seamless, end-to-end process that automates the traditionally manual and error-prone task of interpreting medical prescriptions. This system has significant potential for integration into healthcare management systems, offering a scalable solution for improving prescription processing and patient care.

The rapid digitization of healthcare has led to a growing reliance on technology for storing and managing sensitive patient data, known as electronic health records (EHRs). This increasing dependence on digital systems has heightened concerns about the protection and security of this critical information. With multiple stakeholders involved in healthcare systems, ensuring the integrity and confidentiality of EHRs has become a paramount issue. The work proposes an integrated approach that leverages behavioural security to enhance the security of healthcare data. By tracking the behaviour of users accessing the system based on various parameters, a one-class Support Vector Machine (SVM) model is trained to detect anomalies in user behaviour. If any anomalies are detected, the system is configured to reduce access control for the respective user, effectively mitigating the risk of unauthorized access. This approach has demonstrated positive results in identifying and preventing the unauthorized use of the healthcare system. The implementation of this behavioural security framework, combined with the one-class SVM model, provides a robust and proactive solution to safeguard the confidentiality and integrity of sensitive patient data in the healthcare domain. By continuously monitoring user behaviour and adapting access controls accordingly, this work contributes to the development of more secure and trustworthy healthcare technology ecosystems.

The traditional classroom mode of learning remains highly effective for student learning. However, with the increasing presence of digital gadgets it has increased distractions for students making it challenging for students to maintain their focus levels in the classroom. Also, with larger class sizes teachers also find it difficult to monitor and engage with students effectively. Using Edge AI technology our system works on these issues by providing real-time insights into student activity in the classroom. Running locally on the teacher’s laptop our system uses a containerized AI model to analyze live camera feeds and track attentiveness. Our system allows teachers to input students overall marks which is used to dynamically adjust the threshold level of attention based on past grades and performance. This personalized approach supports each student’s learning journey. By addressing these challenges our system enhances the classroom environment and makes learning more effective.

The efficient functioning of public transportation systems is pivotal for societal connectivity and economic progress, as they serve as lifelines for commuting and mobility. However, the dependency on manual ticketing processes often leads to bottlenecks and inefficiencies, hindering smooth operations and customer satisfaction. Our work focuses on developing an Automated Ticketing System for public transportation, utilizing Computer Vision and Neural Networks. Through the incorporation of Neural Architecture Search and the integration of Deep Sort, a Deep Learning-based object tracking model, we aim to enhance system efficiency. Our study demonstrates promising results, indicating the potential for streamlined ticketing processes in public transportation.