Dharmateja Chelamalasetty

Analyst Intern @Copart

Retrained Deep Neural Network image recognition models for vehicle damage scoring on four corners using CNN and transfer learning, achieving an AUC of 0.89, achieving 20% improvement. • Implemented augmentation, Hyper opt, callbacks, class weights, and gradient clipping techniques to achieve a 30% reduction in overall model loss. Optimized model performance through quantization and parallelization techniques, achieving a 20% runtime improvement. • Enhanced OTL classification model using GBM by integrating scores from Deep NN Model, resulting a 6% AUC improvement. • Implemented new Pro-quote model for the DE team, utilizing GBM to anticipate the price quotation (Estimated Sale Price in Auction) for a Lot, and able to achieve Mean cross-validation score to 88%, with RMSE 1232.25 through data preprocessing, feature engineering, and hyper-parameter tuning which resulted in 2% improvement of existing US PQ Model. • Created Tableau dashboards to offer insights into PQ and Damage score models.