Machine Learning Engineer
SquadRun | July 2017 - June 2019
● Developed Apparel items classification, attribution, and similar image retrieval deep learning pipelines. ○ Utilised Compact bilinear CNN as the base model for all three use cases. ○ The first time that Compact BCNN’s have been employed for solving fashion related use cases and for retrieval tasks. The pipeline performed better than previous state of the art for categorization task. ○ Saved data annotation cost to the company, as pipeline was trained with a paltry dataset of 1200 units per category and incurred zero cost for annotation of bounding box/ clothing landmark coordinates. ● Designed advertisement classification pipelines using text-based deep learning models, contributing to >50% automation for tagging of advertisements for an e-classifieds client with 50k daily SKUs at >95% precision. ● Designed and engineered Speaker Diarization system for SquadVoice, which separates speech intervals of different speakers on an audio recording and then concatenates the same speaker audio recordings together. Utilises deep one-shot learning and agglomerative clustering. ● Developed Keyword Detection System for detecting keywords spoken on call recordings for quality purposes like greetings and closure words detection, using convolutional recurrent neural networks.
New York University (NYU), New York
Computer Science, MS | September 2019 - Present
MS in CS at Courant Institute of Mathematical Sciences (NYU)
JIIT, Noida (Jaypee Institute of Information Technology)
Computer Science & Engineering, B.Tech | August 2013 - August 2017
User has not solved any challenges yet.
User has not participated in any contests yet.
User doesn't have any discussions yet.