Raghu Veer Tatineni

Collaborated with the product development team to research deep learning solutions and develop multiple product pipelines. Integrated cutting edge deep neural networks into Augmented Reality (AR) products with the help of open source technologies, resulting in a 9% increase in conversion rate for the clients. Trained high accuracy keypoint detection models to detect earlobes within a range of 2% for use in AR jewelry products. Designed varied model pipelines for detection, semantic segmentation, etc for skincare applications like acne and spots detection (93% accuracy), eyebags detection, wrinkles detection (0.19 bce loss). Developed classification models to determine a person’s characteristics (hair type, beard type, glasses, age, gender, skin color) from the image, all models achieving accuracy of over 90%. Containerized models using docker and deployed the models as APIs using AWS Lambda and FastAPI, improving response time by 55%. Built a Visual SLAM system proof of concept from scratch consisting of Visual Odometry algorithms.

Strategized the approach to tackle the problem of detecting debris on the road for autonomous driving applications. Developed an improved Unet based model to provide semantic segmentation maps for the 3 classes and detect debris, road. Achieved an accuracy of 94.62% on the test data and obtained useful segmentation maps for other images from the internet.

Relevant Coursework: Linear Algebra, Statistics, Probability, Data Structures and Algorithms, Pattern Recognition and Machine Learning, Computer Vision, Reinforcement Learning