Sep 2020 – Nov 2021 · 1 yr 2 mos · Bengaluru, Karnataka, India

• At Airtel X Labs, I worked on the modelling of handset change intent of 2G users, identification proof fraud detection during customer acquisition, identifying users for personalized customer care intervention for bill payment and representation learning.
• The work involved researching and implementing approaches for building machine learning models for over 350M+ users.

Feb 2019 – Aug 2020 · 1 yr 6 mos · Bengaluru, Karnataka, India

• I worked on ranking, recommendation, representation learning and uncatalogued order classification across Revenue & Growth (POP) and new initiatives at Swiggy. I also researched on recommendation, cross-sell and fraud use cases for Stores and Instamart.
• My work and responsibilities included (but were not limited to) formulating data science solutions from first principles, implementation, monitoring, research in relevant problem statements, mentoring and communicating ideas and results to different stakeholders.

Jan 2016 – Feb 2019 · 3 yrs 1 mo · Bengaluru Area, India

• Personalization
• Drive incremental conversions for clients by personalized treatment of on-domain visitors
• Mining web logs using PIG scripts to extract useful features at page/session level for downstream processing
• Building propensity score models to determine whom to target and personalize their experience based on intent
• A/B testing
• Monitoring of deployed model performance over time and against offline performance achieved during training
• Tools used: PIG, Python, R
• Uplift Modelling with Predictive Personas
• Targeted treatment of visitors who are likely to convert given chat thus avoiding self-serve cannibalization
• Offer treatment on the website based on uplift models for new and repeat visitors
• Determine optimal thresholds for models which maximize recall under given offer rate
• Model performance monitoring using Kibana and comparing online vs offline results for same time period
• Tools used: PIG, Python, R, JavaScript, Nashorn
• Representation Learning for Visitor Journey
• Unsupervised feature extraction from web sessions
• Built dataset of over 10 million web sessions to train session representation models
• Used learned features to train Propensity to Purchase model and benchmarked performance against deployed models
• Tools used: PIG, Python, Word2Vec, PyTorch
• Agent Scheduling
• Optimal allocation and rostering of [24]7 human resource for different clients and geographies
• Formulation of Integer Programming Problem for resource allocation under given demand and capacity constraints
• Find optimal allocation maximizing chat coverage/revenue/profit from chat resolution using branch & bound / cutting plane methods
• Developing optimization tool using RShiny for frontend and GLPK as solver
• Working with operations teams for deployment, training and feedback for quick adoption and improvements
• Tools used: GMPL/GLPK, R, Rshiny

Aug 2014 – Dec 2015 · 1 yr 4 mos · Gurugram, Haryana, India

• Promo Optimization
• Developed Heuristic Hill Climbing Optimization approach to create audience viewership driven campaigns
• Captured audience engagement and behaviour on TV Network by Monte Carlo Simulations
• Used regression to model reach function and incremental reach analysis across time slots
• Developed optimization model for Strategic Planning of episodic and launch promotions to maximize unique viewers
• Conducted post campaign analysis to quantify the effect of campaigns and test hypotheses on conversion of viewers
• Optimized campaigns generated 10% yield while meeting the goals set by the Network
• Tools used: Julia, R, Excel

Mar 2014 – Jul 2014 · 4 mos · Bengaluru, Karnataka, India

• Project Title: Real-time Proxy NPS(Net Promoter Score) Estimation
• Reviewed the algorithm developed by Global Analytics and proposed 3 alternatives for estimating NPS from unstructured data
• SVM classification with dictionary based unigram and bigram features
• SentiWordNet Scoring of reviews followed by classification based on threshold scores and their performance on unseen data
• Sentiment Aggregation of feature level sentiments to determine document sentiment followed by classification
• Behavior of promoters, detractors and passives was also studied

May 2013 – Jul 2013 · 2 mos · Bhadrachalam, Telangana, India

• Product Analytics: Minimization of Complaints using Statistical Data Analysis, Life Testing & Technology Upgradation
• Exploratory analysis of sales and complaints data to identify trends and patterns in failure modes and their occurrences
• Performed Hypothesis Testing on influential product characteristics to identify statistically significant characteristics
• Conducted accelerated life testing at critical process steps to recreate failure modes and hence identify the critical operation
• Suggested implementable methods and measures to quantify and eliminate the defect and improve the product robustness
• Predictive Modeling of Plybond Specifications
• Regression modelling was used to develop relationship between new and old specifications
• Revision of old specifications was done using 97.5% upper CL.

Jan 2013 – Mar 2014 · 1 yr 2 mos

• Responsible for coordinating the placement process in ISI Kolkata for the batch of 2014.

Mar 2011 – Aug 2012 · 1 yr 5 mos · New Delhi

• Edited and published articles and interviews related to start-up companies by IIT alumni & students and other events.
• The website has garnered over one million views and its articles have been linked and referred to by numerous other websites

Jul 2010 – Oct 2010 · 3 mos · Hyderabad, Telangana, India

• Business Analytics
• Sector overviews, performance benchmarking, company comparable analysis etc.

May 2009 – Jul 2009 · 2 mos · Noida, Uttar Pradesh, India

• ~Optimization of the HVAC assembly line Cycle Time at the Thermal Manufacturing Division

Jan 2009 – Jan 2010 · 1 yr · New Delhi Area, India

• ~Led a team of 14 people to complete an animation film, featuring original characters, using the cut-stop motion technique