Mar 2024 – Jul 2025 · 1 yr 4 mos · Bangkok City, Thailand · Hybrid

Apr 2023 – Mar 2024 · 11 mos · Gurugram, Haryana, India · On-site

Oct 2022 – Apr 2023 · 6 mos

• Led and managed the Ads Core Marketplace team which was responsible for all Data
• work around Ads Personalisation and Ads Pricing

Apr 2021 – Sep 2022 · 1 yr 5 mos

• Designed and implemented Gojek Ads' first real-time Ads personalisation engine
• for multiple ad inventories. Achieved very high CTR uplifts resulting in significant impact on Ads revenue.
• Conducted extensive research to conceptualize and build a Reinforcement Learning based Ad Pricing Engine to optimize for revenue and handle supply demand imbalance. Achieved double digit percentage uplift on ads revenue.
• Used Word Embeddings to estimate similarity between different search keywords on GoFood. Used it to improve ad retriever which is based on keyword-restaurant mappings. Significant increase in ad visibility.

Aug 2020 – Apr 2021 · 8 mos

May 2018 – Jul 2020 · 2 yrs 2 mos

• Delivered multiple Data Science projects to large telecom clients
• Automated Ticket Routing
• Built a framework to eliminate manual assignment of teams to repair tickets.
• Used Markov Chains to model the flow of tickets across multiple teams. This can be used for resource planning.
• Used Factorization Machine Models and Decision Trees for ticket routing.
• The approach achieved 75% reduction of ticket assignment task force for a telecom provider in the Middle East.
• Read more about it - https://bit.ly/2Ox2cih
• Network Incident prediction
• Used millions of device alerts/ alarms to predict incidents in RF network of a telecom operator spread across India.
• Used Random Forest classifier for the purpose of prediction and bucketing alarms into different risk segments.
• Achieved an AUC of .82 on PR-curve. Class imbalance of the data was around 1:100.
• Battery Management System Optimization for Electric Vehicles
• Conducted an exhaustive study to understand methods for estimation of State of Health and State of Charge of EV batteries.
• Performed binning of measurements from battery and other sensors. Used these features to estimate SOH of Li-ion batteries.
• Unlike conventional methods, this approach can use features such as acceleration, braking, temperature for SOH prediction.
• Achieved an R-squared score of ~.99 on 4 different batteries.
• Anomaly Detection
• Performed anomaly detection on logs of a mail service provider to detect service degradation.
• Used Gaussian distribution for univariate and multi-variate anomaly detection.

Jul 2016 – Dec 2017 · 1 yr 5 mos

• Demand Forecasting for Optimal Content Scheduling for a large European Cinema Chain
• Led a team to deliver Demand Forecasting system using State-of-the-art Machine Learning techniques
• Demand Drop-off
• Used XGBoost to train a model to forecast next week’s demand of a film using its current week’s demand
• Incorporated various new factors such as temperature, rain, holidays, film genre, etc. in the forecasting
• Demand Distribution
• Used Non-Negative Matrix Factorization to learn extreme inter-day & intraday demand distribution curves
• Unique distribution profiles for each movie can be learned using a linear combination of these extreme curves
• Film Clustering
• Used K-Means Clustering to group films into different clusters based on their attributes and performance
• Evaluated the clusters using Elbow method and within-the-cluster variances of various film attributes
• Demand Cannibalization
• Used Binary Log Linear Regression to model the impact on demand of a show due to other shows of same film
• Trained different models depending upon size of a film and format (2D/ 3D) of the shows
• Personalized Product Recommender – Reviews Insight Generator (RIG) (Opera Open 2016):
• Implemented a solution to generate feature level ratings of a product using its reviews available at various e-commerce website
• Used Stanford NLP Parser and a Graph based Technique to disambiguate sentences into sub-sentences having only one feature
• Performed Sentiment Analysis over these disambiguated sub-sentences to generate user ratings for each feature of a product

Jan 2016 – Jun 2016 · 5 mos

Jul 2014 – Jan 2016 · 1 yr 6 mos

• Worked in the Risk team of Barclays.
• Payment-hold Strategy (Germany Barclaycard):
• Analysed credit card data at various levels such as historical payments, bounces, and delinquency, balance, credit-limit, utilization, etc
• Came up with a strategy which has 20% Footprint and 80% Bad Balance capture as compared to the previous Payment-hold strategy
• Credit Abuse Strategy (Germany Barclaycard):
• Identified a pool of potential high-risk customers using the transactional behavior and other account-level historical information
• This strategy will save ~2.5 – 3 Million Euros every year in impairment cost and will be replicated on other Barclaycard portfolios.