Experience

• 

  Beesphere consultancy services

  May 2018 - Jul 2018

  Machine learning engineer intern

  Worked with a team of data scientists on a python based script using NLP & Machine Learning classifiers which gives you a score how much your product is popular and loved among the social media users.

• 

  Verizon

  Jul 2019 - Jan 2022

  Analytics of sales and discrepancy dataProblem: Provided a usecase for decreasing the discrepancies raised by agents for unsettled commissions.Solution: Collected data from the data source using SQL and analyzed discrepancies using Python’s scipy, plotly, matplotlib etc. to provide statistics about discrepancy’s pattern with its resolution code.Business Impact: Insights helped business in decreasing discrepancies by around 7% month on month.Agent’s retention amount predictionProblem: Figure out amount that can be retained from agent’s salary based on his willingness, interest rate, band, experience, current salary and market id etc to increase reserves and decrease chances of agent’s attrition.Solution: Preprocessed highly skewed data with box cox transformation, imputed missing values using combination of different techniques like using maximum frequency, prediction models, correlated features, and generated new features based on feature importance to perform regression.Business Impact: Helped Verizon business in getting more employees going for retention program as model predicted a comfortable amount figure for agents and this all benefitted with 1.43$ million in a quarter.Verizon Synchrony credit card lead predictionProblem: Identify customers that could show higher intent towards a recommended credit card based on past services, current services, number of activations, total income, region id etc .Solution: Stacked ensemble classifier of xgboost, catboost and lightgbm was used with blending on seeds along with hyper parameter tuning and exposed api using Flask.Business Impact: Improved cross selling by identifying potential customers and giving them personalized offers. Show less Predictive model for BHR device’s replacementProblem: Using BHR device's statistics like alarm, arp, cpu cycle etc. a model for device’s health has to be made.Solution: A classification model for prediction of device’s replacement was made by handling highly imbalanced data using a combination of under sampling and oversampling with SMOTE and data partitioning.Business Impact: Improved customer experience with help of pop up about replacement 89% of timesCommission Calculation system for AgentsProblem: Create database architecture for a new compensation type which is shared compensation which calculates shared commission among agents for transactions done by many agents together.Solution: Built a data pipeline in Pl / Sql starting from flat files till the destination tables using sql loaders, cursors etc. and applied partitioning, indexing for better performance along with other complex logics.Business Impact: Enabled business to calculate shared compensation through the compensation portal instead of calculating them manually. Show less

  • Engr -II Data Science

    Sept 2021 - Jan 2022

  • Engr- I Data Science

    Jul 2019 - Sept 2021
• 

  Nirmaan organization

  Sept 2020 - Nov 2020

  Data scientist

  Analyzed and performed statistical modeling on time series call data of Vidya helpline for various aspects of call like talk time , queue time , number of calls and generated HTML report which provides insights using interactive visualization along with a R based simulation script to verify calculated probabilities.

• 

  Cipla

  Jan 2022 - Oct 2022

  Site lead data scientist

  Global Lighthouse: World Economic Forum• Developed compression optimization model for OEE which got selected for lighthouse.• Presented the data science model to World Economic Forum, CTO and upper management of company where inat compression stage iterations from 10-12 were reduced to 2-3 per batch for setting run.• This model has reduced 80% of change over loss in OEE.Operational Excellence Compression Optimization• Problem: Improve yield and reduce number of iterations of test run in compression stage.• Solution: Reduced number of iterations for Fenofibrate 145mg & Lumartem Dt in compression stage byrecommending the input parameter values of machine using analysis and improved yield by achieving optimalweight of a tablet using regression model on various sub stage.• Business Impact: With 1% improved yield approximately 51,200 tablets can be produced more per year and fromiteration reduction approximately 11lac rupees are saved along with human resource, electricity and time peryear across pan Cipla. Show less

• 

  Amazon

  Oct 2022 - now

  Data scientist