Software Development Engineer

• May 2019 - Present

Project: Anomaly detection Developed in-house a 2-fold architecture hosting a real-time anomaly detection pipeline for detecting neps (cotton entanglements) in raw cotton to improve its quality during the manufacturing of yarn. Research Experience Designed a novel real-time algorithm for nep detection comparing the benefits of YOLO, SSD mobilnet, and faster-RCNN models using IoU (Intersection over Union) metrics. Trained model using a live feed from cameras installed on yarn machines. Deployment Experience Built the anomaly detection pipeline from the ground up and deployed it in the company’s factory. • Configured and deployed multiple global shutter cameras on yarn machines to capture real-time images • Processing over 100GB/day of data through a pipeline • Running our nep detection model on the incoming stream of data and generating predictions in real-time • Storing the predictions along with time stamps on Firebase real-time database using an intermediary AWS server for API calls developed using Flask • Developed a live dashboard in React.JS and Flask for displaying predictions and various metrics in real-time. Impact: - Automated the manual process of nep detections. - The system is helping find faults in the yarn-making machinery each costing 15L-20L, at an earlier stage-giving abnormal data. Ongoing Project: Commodity Price Prediction Team Role: Lead Designed the layout and worked on the development of a commodity(cotton) price prediction tool. The aim is to automate the process and also make accurate predictions of the price of cotton in the medium term. Worked on different data sources to acquire relevant data and thereafter apply CV algorithms to generate results. Major data sources for analytics include: • Satellite Imagery • Social Media Analysis • RSS Feed data • Chart Data Analysis Show less