Recommendations Platform Representations - Built and grew to a team of 12 MLEs across multiple timezones - Responsible for common user interest based representations, and the embedding-based candidate generation for multiple product verticals. Collaborated with researchers and product teams to improve consumer recommendation - Leveraged techniques from Neural network, GraphML, big data processing

Recommendation systems for consumer products - Two-tower model-based tweet similarity - Topic tweets recommendation

Pure1 Meta Workload Planner: We utilize machine learning to develop a planning tool that allows customers to predict their storage array loads, with different hardware configuration or workload allocation. This features marks as the #1 page views on Pure1 (Pure Storage's cloud-based management tool) besides the landing page. - Model research (data cleaning, model design/selection/validation, hyperparameter tuning) and define machine learning metrics to hit the product goals. - Researched and utilized machine learning models such as random forest, gradient boosting, KNN, MLP, LSTM, etc. - Implemented ETL pipelines in Spark to extract features and reference data. - Productionize models in REST servers. - Designed and the implemented automated testing/experiment pipelines to speed-up model research. - Lead the effort to help the data generation team to produce cleaner data. - Interact with sales engineers and customers to understand product requirements and expectations. Leadership: - Co-founded study group within the Pure1 business organization. This is a recurring event that allows engineers and PMs to learn from others through hands-on labs. - Co-hosted super lightning talks events in which teams pitched their projects in 10 slides and 3 min 20 sec. Award: - Archimedes Cup (Pure1 best individual of the quarter), award to one of 150+ employees within Pure1 business organization - Hackathon Creativity Award Pure1 META: Pure’s AI Platform to enable Self-Driving Storage

Cloud-Connector project: Drove and implemented in Java a new log uploading module that replaced the legacy system. The new log uploader utilized in-memory databases to improve efficiency and robustness. Contributed to the design for modularization and multi-device support.

Led weekly discussion sessions, discussed with instructors to develop course materials and quizzes, and evaluated students through grading.

Led weekly discussions, and guided students in a magnetic levitation project where they designed PID controllers that levitate a metallic ball by utilizing Matlab and Simulink.