NEERAJ KUMAR

Buffer Forecasting for leading Telecommunications industry

Technology/Skills: Python, Time Series Forecasting, Machine Learning, Deep Learning ,pyspark, Azure databricks

Gap—To Predict buffer stocks of telecommunications equipment & Scrap reduction (Inventory Optimization)

Leading a Data Science team and built Safety Stock Prediction model ( part of Demand Forecast) for Global Telecommunication giant .

Using ML, Deep learning .Traditional Time Series modelling.

Impact 1: reduced Scrap by 14% in 5 global hubs, estimated cost savings $ 112 M annually

Impact: 2: Stockouts Prevention for 9% of total SKU units

Sales Demand Forecasting

Technology/Skills: Python, Time Series Forecasting, Machine Learning, Deep Learning , pyspark ,Tableau

Gap—To Predict Sales Volume for leading Indian Retail Sector Company

• Time Series/Regression Modelling for Sales Demand forecasting at pan country level

• Used scalable ML modelling concepts :Like Time Series Clustering ,Pooled ML models, cross validation techniques

• Time series features, ts learn ,XGB , RF, Neural Networks , Fbprophet, ETS, Arima

Tools :Python , pyspark , spark , Microsoft

impact – Increased the sells for Priority 1-brands by 14 % quarterly and optimized cost effectiveness in supply chain by 4.3 % overall (Estimated cost saving $ 13 M annually)

Virtual Machine Crash Alert model – Premium Customers

Technology/Skills: Python, NLP ,NoSQL ,Tableau , Linux

Gap – Flagging Premium Customers VM’s IOPS if crosses threshold weekly with Time Series/Regression Model

Solution –

Solved Business Challenges via Predictive Analytics & NLP on improving the threshold of customers VM usage.

Using time series and regression classified the customers who has critical usage and unusual future trends. [VM Crash Alert Model] with 76 % accuracy to increase their VM IOPs limits to avoid crash

Automated the process weekly with flagged critical customer VM usage .

Global Ticket Management Model

Technology/Skills: Python ,R, NLP, Deep Learning , SQL, Tableau, Linux

Gap - Due to heavy manual intervention in assigning the right ticket to the right group, the support team had huge backlogs and a very high ticket resolution time

Solution –

Using the text in the description (along with service category) of the ticket, created a document term matrix as the input to a Random Forest Model. First Model accurately categorizes the ticket type into service request or service restoration with 86.7% accuracy over 100K incidents (tickets) quarterly

Once successfully categorized, it assigns the ticket to 1 of 60 resolution teams with an accuracy of 73.2%

Automated the above two models to notify the respective groups about the open tickets every 24 hours using Cron Scheduling. Results - Saved over 500 man hours for ticket assignment & reduced the average ticket resolution time from 72 to 52 hours quarterly

Fraud Management Pipeline Models & Analysis

Technology/Skills: Python ,R, NLP, Deep Learning ,Hive, pig , Tableau, Linux,

Solved Business Challenges via ML/ Text Analytics & NLP on reimbursement receipts to improve audit processes & maintain company compliance policies. Some major projects include:

o Extraction of line items from Hotel receipts

o Classifying model for meal receipts with/without alcoholic beverages

o Predicting number of attendees model in a meal receipt

o Identifying Fraud in mileage reimbursement

· Impact: Saved over $120K annually in reimbursement compliance breach

Technology/Skills: Core java, HTML,CSS

Assisted in maintaining the landing page of the primary website and in fixing the webpage bugs related to padding, width and other UI fixes