* Building and productionzing solution for RTO using Logistic and Boosting algorithm. (Return to Origin (RTO) is any shipment which is being returned to the origin address of the shipper due to Non-delivery because of underlying reasons.)* Proven 9% reduction in RTO rate on 2 merchants (Monthly GMV pliftment by 5.5 Lakh INR)* Built Global RTO model to handle 20+ merchants with 9% better precision than there Bespoke version on average.* Participated in inhouse Kwikathon and won the 1st prize for building chatbot using RASA.

* Setting up in house ML Ops Capabilities for Data Science (Model (re)training pipeline, Model Monitoring System etc.)* Building ETL pipeline using PySpark, Postgres, Python and Bash scripts.* reduced bert model latency by using by introducing TensorRT. (https://medium.com/@niteshsukhwani97/improve-inference-time-for-tensorflow-models-using-tensorrt-8d3a1c3b0ada)* Handled custom model deployment using Vertax AI (GCP)

IDCC (Intelligent Data Capture and Control): The main objective of the project is to extractstructured information from metal plates of assets of manufacturing plants. My role is to apply Machine Learning/ Deep Learning techniques to make the product more efficient and generic.

Alternate Vendor & Part: Use web scraping and crawling from selenium to make a generic scraper for extracting details of multiple vendors of a given item and also provide substitutes for that item using similarity in feature from the base product

MBD-417 Python and JavaDuties: Leading discussion sessions, mentoring students in programming concepts.

This was a research internship where I have worked on the problem of finding efficiently skyline probability on a dataset having uncertain preferences and finding top K skyline object.