Ajay Singh

Experience

• 

  HCL Technologies

  Jun 2011 - Jul 2011

  Software Trainee
• 

  BSES Delhi

  Jun 2012 - Jul 2012

  Software Trainee

  Developed a Flight Management System in ASP.NET with C# as my summer Internship.

• 

  PayU India

  Aug 2013 - Feb 2015

  Software Engineer

  Part of the development team of New Merchant Panel for PayU's merchants to monitor/analyse their transactions traffic and display result using pie charts and graphs.Integrated banks with payment gateway to facilitate direct transactions while strengthening the end-to-end security of its Internet banking services.Developed various web services and customization options for payment flow for merchants.

• 

  Delhivery

  Mar 2015 - Feb 2020

  · Built a system to predict the path taken by a bike using mobile GPS data. Used ML algorithms to resolve GPS issues, e.g. missing GPS, irregular GPS points, U-turn behavior, etc. This path is being used in Automatic Fuel Payout of delivery agents and helped in saving INR 10M per month.· Trucks ETA Prediction - Developed and deployed an algorithm basis RandomForest to predict the ETA of a Truck using GPS data. Finally, missing roads found during this were contributed to OpenStreetMap.· Address Type Prediction - Implemented SVM and Naive Bayes to classify houses as residential or commercial using their address.· Locality Polygons - Created geographical boundaries of cities and localities using DBSCAN, OSRM and tagged geopoint data with roads as solution space. Show less Converted Delhivery’s operations lifecycle into a time-based network graph and worked on suggesting shortest paths, alternate paths, reducing misroutes, etc.Generic-wrapper: Developed a Queue-based API service architecture to serve millions of requests to time-consuming ML algorithms.Naksha: Developed an algorithm to derive intelligence, eg. geofencing, routing etc. from GPS data and developed microservices around it. These polygons were later stored in the address hierarchy of state, city, locality, and sub-locality to assist in geocoding and reverse-geocoding process.Catalog-UPI: Collected data from many e-commerce sources. Tweaked scoring mechanism of ElasticSearch to find the closest match of a new product. Created flow for manual verification of all attributes of a product to be able to uniquely identify products from an internally developed category tree.Developed various internal tools. e.g. Locality Management, Cataloging, Report Central. Show less

  • Data Scientist

    Apr 2016 - Feb 2020

  • Developer

    Mar 2015 - Mar 2016
• 

  Bajaj Finserv

  Feb 2020 - now

  · Personal Loan Propensity Model - Implemented an XGBoost model utilizing the app, clickstream, and bureau data to enhance explicit lead conversion rates, elevating them from 4% to 12% for PLCS· SALPL Lookalike customers - Utilized contact details, email domains, EPFO data, and appography toidentify salaried customers for PL, resulting in an additional 155 crore revenue on the app and web platforms. Simultaneously, achieved a reduction in the Cost of Acquisition (COA) from 4.5% to 3.5% per disbursal· Developed and deployed highly accurate facial recognition algorithms (including VGG19 and RESNET),reducing manual document verification process costs by 40% Show less · Deployed a DBSCAN-based GPS clustering algorithm to pinpoint customers’ current locations, sub-sequently refining the clusters with CPV and receipts data. This initiative has resulted in several impactfuloutcomes, including Skip Tracing of delinquent customers ( 15Cr saving), a 9-12% reduction in Match Rates for RCU [for Ecom], enhanced lead mapping of customers to Field Officers, and resolution of PL-OGL casesto generate an additional 150 Cr, among other benefits· Customer’s Primary Number Tagging - Designed a name prediction algorithm based on phonetics, lever-aging the contacts data within the app to accurately identify primary and secondary contact numbers forengaging with BFL’s prospects. This initiative significantly reduced wrong number (WN) tagging and in-creased the success rate of Right Party Calling (RPC) from 76% to 91%.· Risk Analytics B2B: Conducted multiple Design of Experiments (DOEs) utilizing CHAID analysis to tran-sition approx 79L customers from New to Bajaj (NTB) to Prospects to Bajaj (PTB) offer pool for B2B Show less

  • Principal Data Scientist

    Apr 2023 - now

  • Senior Lead - Data Science

    Oct 2021 - Apr 2023

  • Lead - Data Science

    Feb 2020 - Sept 2021