India

Master's degree

Bachelor of Technology - BTech (Honours)

Shaurya Uppal

Sr. Data Scientist

I worked on two Products-Roposo (Short Video Entertainment and Commerce Platform) and Glance (LockScreen Entertainment and Awareness Platform).

Glance Roposo is a short-video entertainment and commerce platform.
▪️ Tier 2 Creators Persona Identification and Churn Prediction using K-Prototypes Clustering Algorithm on Creator Demographics Data and Engagement Activity.

▪️ User-Creator affinity identification using multi-criteria decision-making technique Topsis and graph algorithms (node2vec) on profile visit, follow, like, etc. This helped the platform provide a better recommendation for creator stories to users where user-creator affinity is high.

▪️ Gender Prediction: Imbalance Classification problem on behavioral features of the user; the majority class is male and the minority is female. Best Results: LightGBM + Undersampling technique EditedNearestNeighbour resulting in an AUC of 0.73 and an F1 Score of 0.69

▪️ Live Stream Ranking of Commerce Product - Based on User-Creator Affinity computed using the user's past interactions with the creator.

Glance Monetisation 
▪️ For Ads targeting on LockScreen for different brands: finding Look-Alike User Segments
▪️Identify Similarity between Apps based on User Behavior of Installs - Similar App Recommendation and Ads Personalisation
▪️ Capturing user interest based on app installation, improved one-click-install (OCI), Conversion per Mile, and LTV Revenue actions like Registration, Purchase rate across Advertisers by 20% 
▪️Ads personalization for better targeting and maximizing revenue goals: A/B Testing Metrics: 16% CVM improvement and 38% LTV improvement. 
▪️Creating personalized targetting user segments for Voucher Procurement Campaigns. CTR improved by 18% (A/B testing comparison)
▪️Brand Ads Targeting for better ROI. Improving CTR by 18% and Conversion of Installs and Registration by 10%

⭐️ Awarded Bond007 title at InMobi. 
⭐️ Application for Patent on Ads Personalisation

Worked on Ranking Problems, Real-Time Rating Prediction of a Product, Chances of a Loan Approval for a User based on Product-User Features. 

Home-Page Personalisation based on User's eligibility and financial survey scores.

Neobanking Startup by Google Pay Founders backed by Sequoia Capital, Ribbit, and other giant VCs.

▪ I joined here as a part of the top@50 early members of the organization and second in the data science team. I built core features and structured the data science team, making the new-generation millennial banking experience more personalized and user-friendly.

▪ Setup MLOps components: Airflow for Automated Model Pipelines, Data Version Control, BentoML for Model Inferencing system, Kibana for Logging

▪ Data stores: SparkSQL(data lake); RDBMS: PostgreSQL, CockroachDB, Snowflake

Responsibilities:
▪ Hiring the best talent for the company
▪ Mentoring Interns and Freshers in Data Science team
▪ Business requirements understanding, communicating metrics and helping leadership team in making data-driven decisions

Problems Solved:
▪ Ranking of Users on Fi Waitlist using Multi-Criteria Decision Making 
▪ Retaining Users on Waitlist by sighting them with our feature Insights powered by NLP Data Science Model. Here we look up the user's merchant mails and extract price using a NER Model.
▪ Understanding customer support requirements and helping customer support agents by automating the issue categorization process. Issue categorization helps us know the friction points of our product. Powered with Zero-Shot-Classifier Model and list of predefined issue tags.
▪ Improving in-app help using data by making changes to the help screen so one can find answers faster. To make it easier to find the right answer after landing on the wrong search result, added 'Related articles.' such that related articles are semantically similar to the one you are currently reading. Then this entire flow of finding related articles is automated by Airflow.
▪ Based on user features Demographics, Geographics, Technographics Features, Income, etc., group users can find accurate user peer groups using the K-Prototypes clustering Model. Outcome: Over this peer group show personalized user insights.

ProjectPro (Formerly DeZyre)

Subject Matter Expert - Data Scientist (Consultant)
Created top-notch research-driven quality courses, which show my ability to study, learn, communicate, carry out research in a subject, and representation skills in the Data Science field. 

My published courses are below:

[1] Ecommerce product reviews - Pairwise ranking and sentiment analysis
URL: https://www.projectpro.io/project-use-case/ecommerce-product-reviews-ranking-sentiment-analysis

[2] Demand prediction of driver availability using multistep time series analysis
URL: https://www.projectpro.io/project-use-case/demand-prediction-time-series

[3] Ola Ride Request Demand Forecast 
URL: https://www.projectpro.io/project-use-case/ola-bike-rides-request-demand-forecast

[4] Building a Graph Based Recommendation System: https://www.projectpro.io/project-use-case/graph-based-recommendation-system-python

[5] Deepwalk and Node2vec Graph Recommendation at E-commerce: https://www.projectpro.io/project-use-case/graph-based-recommendation-system-python-for-ecommerce

I have worked on some complex and interesting industrial problems.

▪ On product pages of 1mg, we showed a Similar Products List which was to improve platform diversification by allowing a user's to explore and exploit the platform. This similar product widget was powered by BERT-word embedding on the meta-data of the product and the deepwalk prod2vec technique.

▪ Personalized Recommendation of Products using Neural Collaborative Filtering, this personalized recommended list in comparison to the Popular Products list increased Add To Cart by 32% and CTR by 38%.

▪  Worked on the entire homepage metric pipeline from setting up CTAS queries on Athena to Scheduling Script on Airflow with a flock webhook that posts daily metrics. 

▪  Worked on Trending Products Widget on Search, which is on ZScore Moving Average idealogy. It helped us gain 60% orders by this widget when compared in the A/B test with a popular product list.

▪  Finding Disease Trends time series analysis.

▪  Ranking reviews by relevance: sorting reviews by relevance.

▪  Worked on search algorithm enhancement, reduced unsupported search by 20% using NLP methods, LDA, TextRank, and TF-IDF on product descriptions to extract related keywords and phrases.

▪  Worked on Product Filter Tags automatic generation for products that helps customers to easily find a product on our page.

▪  Multilabel Classification, Auto Categorization of Products to different Categorizes using product features like color, name, uses, product form, ingredients, etc.

▪  Creating an NLP library that has the ability to pre-process data, extract features, do phonetic matching, string matching, get readability score, etc.

▪  Developed content-based similar products model using NLP methods, word2vec, dimensionality reduction, TF-IDF, BERT, and similarity algorithms.

▪  Awarded First Data Science Sprint Hero @1MG

InterviewBit is an E-learning Platform (Top #10 startups of India by LinkedIn) which also conducts hiring challenges. 
For Machine Learning/ Data Science challenges, I have been their problem setter. 
This experience has taught me a lot, I get to work on different problems like Text Classification, Text Summarization, Keyword Extraction from Text, Regression and Classification problems. 

So far, I have worked on around 14+ problems with InterviewBit which were used in Pan India competitions of which in one of them, more than 200k+ developers participated.

Shaurya Uppal🇮🇳

About Shaurya Uppal

Expertise

Sessions with Shaurya Uppal