Dhushyanth Polkampalli

Developed a hybrid (classification + regression) deep learning algorithm to predict product pose (horizontal angle) in
Wayfair images: Achieved an accuracy of 74.6% for 8-Zone angle prediction and accuracy of 42% in top-22.5° for angle
● Created Wayfair imagery datasets for identified high-revenue and high-volume Wayfair products
● Augmented training dataset by generating synthetic data using 3D models of selected products by scripting in 3DS MAX
● Reduced domain gap in synthetic images by improving image composition: using Gaussian noise and histogram matching

• Designed the Call Analyzer data pipeline by leveraging Python multiprocessing and Spark distributed processing, achieving a 10x speed improvement.
• Conducted detailed profiling of AWS Spark Glue jobs to identify performance bottlenecks and implement enhancements.
• Utilized SageMaker deployments to parallelize GPU-accelerated computations for efficient embedding generation.
• Engineered sophisticated ML solutions integrating Vector Databases and Large Language Models (LLMs).
• Developed a customized solution using modifications to Retrieval-Augmented Generation (RAG) for customers with LLM/GPT restrictions.
• Designed and deployed end-to-end data processing pipelines for various clients of Call Analyzer on AWS.
• Architected the Call Analyzer platform, encompassing comprehensive design, documentation, code standards, deployments, and DevOps practices.