Worked in Search relevance team.

• Developed a end-to-end ML pipeline that includes feature engineering, training GBDT model & A/B testing for query category prediction.

• Designed and implemented an active learning strategy to identify incorrectly labeled data and retrain the model with modified weights to avoid the need for remarking.

• Developed E-commerce product classification end to end pipeline which uses CatBoost and BERT embeddings related cross features.

• Developed Miscategorized Item Detection algorithm using FAISS's approximate nearest neighbor and heuristics based on price, user behavior.

Worked on building NLP models and predictive modelling

• Survival Modelling using XGBoost classifier using patient historical data.

• Build Smoking status classifier by finetuning pretrained ULMFIT model(transfer learning) using limited number of labelled data.

• Build regex based pipeline to extract different attrbutes using patient's report.

• Classification of different e-commerce products into product types generated from taxonomy using SVM. Successfully delivered this product output to 20 client companies and got appreciation from the CEO.

• Attribute value extraction from the e-commerce data using BiLSTM-CRF NER model.

• Product Matching in E-Commerce: Developed text based deep learning model for finding similar products across thousands of e-commerce stores containing millions of products. Text-based mapping includes the classification of products based on their normalized attribute features.

• Build Global Ecommerce knowledge graph for dataweave using Google and GPC taxonomy.