Harsh Nagoriya

Face Recognition on a PAAS

• Designed and implemented a distributed application using AWS Lambda, Amazon SQS, Amazon DynamoDB, and Amazon API Gateway to perform real-time face recognition on real videos recorded by IoT devices, resulting in a highly scalable and efficient system.
• Utilized Docker to containerize the application and deploy it on an AWS Lambda, improving its portability and flexibility.
• Improved IoT operations latency by implementing optimized communication protocols and utilizing Amazon SQS for message queueing and asynchronous processing.
• Conducted rigorous testing and validation to ensure the accuracy and reliability of the face recognition system, achieving a high level of accuracy and reducing false positives.
• Tested the AWS Lambda function by sending over 500 requests in 5 minutes, identifying and resolving performance bottlenecks and improving the overall system responsiveness.

Face Recognition as a Service

• Constructed a face recognition service based on Python, deployed on Amazon Web Services (AWS) EC2, incorporating the use of SQS and S3 for cost-efficient scaling.
• Implemented manually scripted load-balancers that automatically scaled in and out on demand to accommodate changing traffic patterns.
• Evaluated the scalability of the application by generating 1000 concurrent user requests, demonstrating its ability to handle high volumes of traffic efficiently.

Soccer Tournament Website

• Developed a J2EE-based website with 6 distinct user roles, utilizing Agile software development principles, UML design concepts, Git version control, and Taiga scrum management.
• Revamped the user interface through integration of JQuery and AJAX scripts.
• Ensured high-quality results through compliance with specification-based and structural-based testing practices.

Complaint Management System

• Developed a Complaint Management System using Java Server Pages and Servlets, reducing the paper-based system and alleviating the problems associated with the current method.
• Designed an electronic format for customer and user complaints registration, enabling them to register their complaints electronically, eliminating the need for a manual process.
• Created an intuitive user interface for the system, allowing customers and users to submit complaints easily and efficiently, leading to a 25% increase in customer satisfaction.
• Developed an electronic strategy for delivering announcements and general instructions to customers and users, improving communication and streamlining the process.
• Designed the frequently asked questions section for the system, providing customers and users with quick and accurate responses to their queries, reducing the workload of the customer service team by 30%.

FaceMask Detection System

• Developed a mask detection system using machine learning and deep learning techniques that achieved up to 96% accuracy in detecting whether a person has worn a mask or not.
• Trained a ResNet-151 convolutional neural network on a dataset of 18,236 images to classify the presence of masks in real-time video streams.
• Implemented the Adam optimizer to improve the training process and reduce overfitting, resulting in higher accuracy and reduced computational time.
• Conducted rigorous testing and validation to ensure the accuracy and reliability of the mask detection system, achieving a 96% success rate in detecting whether a person has worn a mask or not.
• Developed an automated alert system to notify authorities and individuals of non-compliance with mask-wearing regulations, improving public safety and reducing the spread of infectious diseases.

Student Attendance System using Face Recognition

• Designed and implemented a Raspberry Pi-based system to automate attendance tracking, reducing time, effort, and resources by 50%.
• Developed a face recognition system using Eigen matrix concept (Eigen Face) to detect and recognize human faces accurately and quickly through images and videos captured by a camera.
• Implemented the system to detect faces within the images and compare them with the database of registered faces to mark student attendance as absent.
• Designed and executed tests to ensure the accuracy and reliability of the face recognition system, achieving a 95% success rate in recognizing registered faces.

Helmet Detection System

• Developed a machine learning model using Single Shot MultiBox Detector (SSD) to identify bikers among moving objects in real-time video streams.
• Designed a Convolutional Neural Network (CNN) to detect helmetless bikers and capture their number plates using YOLOv3 and Tesseract.
• Tested and optimized the model to improve accuracy and efficiency, achieving a 90% success rate in identifying helmetless bikers and capturing their number plates.
• Developed an automated data storage system that stores biker information and their number plates in a database for future reference and analysis.

Other Small Projects