Project 1: Title Processing

◦ Led a team of 2 ML engineers & 2 software developers to automate the circadian work of vehicle titles processing team

◦ Eliminated ambiguity by outlining every minute detail in the Technical Specification such as product’s scalability, objectives of business and functional requirements from stakeholders in a precise manner

◦ Achieved 2x throughput from the titles team by developing a ML pipeline which would use image processing techniques like Object Detection and Optical Character Recognition to automate the Titles Processing task

◦ Induced modularization and fast deployments by opting microservices architecture

Project 2: Audio Event Detection

◦ Detected knock/ticks from engine audio recordings by performing investigatory analysis of large audio datasets

◦ Successfully developed an ensemble of deep learning and XGBoost based working prototype of engine noise detector by leveraging some of the latest advancements in the audio event detection domain

◦ Presented a viable end to end solution to the CTO and stakeholders for curbing the losses incurred due to arbitration

Project 3: Data Generation Policy

◦ Developed methodologies to identify high-quality data and took an initiative to update the data generation policies company wide by establishing best practices

Project: Price prediction

◦ Increased profitability by reducing the pricing team requirement from 10 FTE to 2 FTE and achieved a MAE of $500 by developing the best in class pricing model for used cars

◦ Achieved 30% higher acceptance of recommended price and 9% increase in the auction success rate by performing in-depth analysis of historical auction data and using it in model calibration◦Identified and constructed informative features by performing exploratory analysis on vehicle and market attributes

◦ Solved the problem of cold start by developing effective feature embeddings◦Increased transparency by developing dashboards which would track performance of the model in real-time

◦ Achieved reliability and confidence by automating the entire workflow starting from training and testing the model with new data to the deployment process