Praty

Business Research Analyst I

Amazon

May 2021 - Sep 2023 • 2 yrs 4 mos

Built a text-based clustering pipeline using Sentence-BERT and T5 LLM to automatically identify topics in customer questions at multiple levels of granularity. Reduced taxonomy creation and data-labeling time from 2 weeks to 2 hours, with scalability across 600+ product types. Deployed ML/DL models on AWS SageMaker, leveraging distributed computing for large-scale training and inference across diverse retail datasets. Productionized NLP models to extract product attributes from unstructured text sources such as titles, bullet points, and descriptions. Successfully deployed across 135 product types and 8 marketplaces, contributing to an estimated business impact of $640M. Developed and deployed a computer-vision–based image similarity model to identify and eliminate dissimilar images in the product catalog. Implemented across 37 product types, resulting in the removal of dissimilar images from 11% of the catalog and improving catalog quality at scale.