FallGuard is a fall detection system built to overcome key limitations of vision-based approaches, such as privacy concerns and camera blind spots. It leverages wearable sensor data to provide accurate and real-time fall detection, offering a reliable alternative in both indoor and outdoor settings.

The system utilizes triaxial accelerometer signals from wearable devices, augmented using SMOTE to address class imbalance. It employs an enhanced InceptionTimePlus architecture tailored for time series classification, capturing subtle temporal dynamics associated with fall events.

FallGuard achieved a classification accuracy of 98.84% and a fooling rate of 96.29%, outperforming four established deep learning baselines. Its strong performance and generalizability make it well-suited for integration into healthcare monitoring systems.

GaitSense is a drift-aware classification system designed to identify temporal distribution shifts in gait signals, particularly those linked to freezing episodes in Parkinsonian movement. It addresses a significant 43.35% classification accuracy drop typically caused by such shifts in time series data.

The model architecture integrates Principal Component Analysis (PCA) for dimensionality reduction, SMOTE for handling class imbalance, and Temporal Convolutional Networks (TCNs) to classify drifts detected using a Bayesian change detector. The input comprises triaxial accelerometer signals, allowing fine-grained temporal pattern recognition.

Tested on the Daphnet Freezing of Gait dataset, GaitSense outperformed three state-of-the-art drift detectors, achieving 96.9% classification accuracy and a 92.3% fooling rate, demonstrating its robustness in dynamic and high-noise environments.

NavAid is an assistive navigation system designed to support visually impaired users by detecting up to 91 objects and providing real-time voice guidance along with GPS-based location sharing. The system enables users to navigate their surroundings safely and independently, both indoors and outdoors.

It is developed using YOLO for general object detection, Haar cascades for face recognition, pyttsx3 for voice output, and LangChain for enabling offline navigation logic. Integrated GPS sharing further enhances user safety by allowing real-time tracking and assistance when needed.

The entire solution is deployed on a Raspberry Pi, achieving a detection accuracy of 98.63% in live conditions. Real-time performance and offline capability make NavAid a reliable companion for enhancing mobility and accessibility among the visually impaired.

I contributed to this project by hardware procurement and assembly, Documentation, Configuring raspberry PI.

For more details click View Code

Streamlining railway operations through smart technology: This web application was developed to efficiently manage train schedules and over 100 booking records with a focus on security and usability. The platform ensures authenticated access to sensitive operations and ticketing functions, tailored for administrative control and user convenience.

The system leverages Flask for backend routing and session management, MySQL for structured data storage, and HTML, CSS, and JavaScript for a responsive front-end interface. Secure transactions and login processes were reinforced using Triple DES (TDES) encryption. The backend logic followed Object-Oriented Programming principles, enabling scalable and maintainable code.

Key features include automated PDF ticket generation for every booking and real-time voice alerts to guide users through the interface, both of which enhanced overall usability. During testing, the solution performed successfully across 90% of scenarios, demonstrating its reliability in real-world conditions.

I played as a team leader, implementing the overall backend services and TDES (Triple DES) implementation.

For more details click View Code

Docker is Designed to make applications to be OS independent, portable and easy deployable into containers. Users can just pull the image from registry and work on their host machine. Kubernetes helps us to scale the pods up/down according to user traffic.

This project "Containerization using Docker and load scaling using Kubernetes" offers a lightweight, efficient, scalable, and highly portable alternative to traditional VMs, making it a powerful solution for modern cloud-native applications.

I contributed to this project by implementing an application using PHP for Containerization and Load Scaling.

For more details click View Code