At VIT Chennai, I prepared 10,000+ RGB-D samples for machine-learning-based pose estimation and compared regression models, reaching 76.5% accuracy within defined error thresholds. I also reduced model loss by 78% through hyperparameter tuning.
For my Customer Behavior Analysis project, I used SQL and Python to analyze 5,000+ retail transactions, then developed Power BI dashboards covering customer segments, purchase frequency, revenue trends, and product performance.
I also worked on an image-detection system for a BeetleBot-based crack-inspection project, integrating ROS 2, camera input, and image-mask generation. I benchmarked three FP16 segmentation models, with EfficientNet-B0 reaching 11.56 ms inference and MixNet-S achieving 0.8180 mean IoU.
