Tarun Aditya is a data science leader with over 15 years of experience leading teams and building predictive models and data products for clients in various industries including ad tech, ecommerce, retail banking, and more. He has expertise in machine learning, deep learning, data engineering, and cloud platforms. Currently he is an Associate Director at Affle where he manages a $50M product portfolio and has introduced neural network models that improved key metrics like ROAS by 38%.
Tarun Aditya's Data Science and Analytics Experience
1. Tarun Aditya (Kaggler, PMP, Six Sigma Black Belt)
https://github.com/tarunaditya
https://www.kaggle.com/tarunad
✆ +91 8884534666 Skype: tarunaditya27
✉tarunadityab@gmail.com
I ENJOY LEADING SMART MINDS IN TRANSLATING LONG TERM STRATEGY INTO TACTICAL DATA SCIENCE +TECH SOLUTIONS & OPERATIONAL SUPPORT
● LEADING DATA SCIENCES PRODUCTS TO BUILD USER RESPONSE SYSTEMS FOR AD TECH, ECOMMERCE, RETAIL BANKING CLIENTS (AVG MAU: 100MN)
● MLOPS: FEATURE ENGINEERING/NN ARCHITECTURE EXPERIMENTATION, PLANNING. STABILIZING AUC & DATA DRIFT MGMT, DOCKERIZATION
● MANAGED TEAM’S OKRS & PLAYBOOK; PRIORITIZED & ALIGNED PROJECTED REVENUE(DUE TO MODEL ENHANCEMENTS) & TEAM’S INFRA COSTS
● HANDSON: CODER IN PYTHON, SPARK, TENSORFLOW, JAVA, SCALA, AEROSPIKE,KAFKA; INFRA:AWS, GCP
Associate Director Data Sciences Affle (Mobile Ads, User Experience platforms) [Mar’20-Present]
Managed $50 Mn worth Product portfolio that supports AdTech & MarTech offerings via User Acquisition(UA), Retargeting users in eComm
- Introduced NN based CPI, CPA, CTR models that predicts App Installers, ReEngages users from RTB sources thus lifting ROAS by 38%
- MLP based Tensorflow model gets model queried by C++ based RTB-DSP with latency of 10ms per request
- Improving model adoption & AUC version by version by planning architectural & feature engineering experiments
- Automated Fraud/anomaly detection in affiliate marketing business, blocking 30% of fraud that's estimated to increase client budgets by 45%
- Hosting the isolation forest based model via AWS lambda that filters bots/fraud apps @ 6K qps
- Built several predictive layers (like demographic tags, geo-spatial affinities etc) for DMP platform which led to commercial success in presales
- Tracked Experiments of NN architectures & User context Feature embeddings like Topic models, User query intents via Mlflow, tensorboard
- Preprocessing & Semantic matching of products across advertisers based on structured & unstructured descriptions of product
- Extracting user behavioural data, competitive pricing index & other abstracted features that drives conversions
- Smart notifications in mobile app: Reinforcement Learning framework to build customer engagement platform to boost re-engagement KPIs
- Beats human guessworks in app notification campaigns by 10K user events (26% better)
- System explores the right timing to engage users based on the context & history of user engagement with right message based on interests
- Led data science development through OKRs, partnering with Data engineers, Machine learning engineers, EMs, product managers
Staff Software engineer, Machine Learning Walmart Labs [Oct’19-Jan’20]
- Auto Data Profiler & Anomaly detector: Built an unsupervised profiler to assist analysts, merchandisers & detect anomalies in their databases.
- Topic Modelling on customer feedback at contact center & returns to reduce walmart associates manual store planning effort by 10 man-days
- Improved Mape by 30% for ‘Forecasting products’ by building scraper that gathers covariate data like weather data, IOT, geospatial data etc
Head of Data Sciences Moengage(Building Data Science products) [Jan’19-Oct’19]
- Managed $20Mn Data science platform, processing 10 Bn events/day, Notifying >100mn customers a day with a team of Analysts, DS & DE
- Increased user retention by 8% via ‘Personalized Notifier’: Reinforcement Learning system that adaptively finds ‘Best Time to InApp notify’
- Productized A/B content testing in Email/SMS campaigns & Topic Modelling of content sentiment
- Increased offer consumption by 12% by building Recommender system for Retail Bankers that recommends next best engagement-action
- Networking & Selling organizational offering in various data science conferences & Webinars.
- Inculcating best practices of software engineering practices on AWS, GCP & Engineering practices in Data sciences
Head of Product Solutions, Applied Data Sciences Near.co(Data Products/Mobile-app ad campaigns) [Oct’17–Jan’19]
Developing & Building multiple dimensions (namely Data sciences, Data Engineering, Product Solutions, POCs & Account management) of
product building. I contribute to building 'Geo Location Intelligence' platform by ingesting > million RTB bids per minute & Telecom data,
curating & maintaining targetable audiences for Mobile ad campaigns.
Built custom decision support data pipelines for eCommerce players based on competitive intelligence & offline intelligence
- Deploying production models in Demographic & behavioural profiles (gender, age, segments etc) at scale & pipelines that handle 0.2 mn Qps
- Enriching places data, mobile users data with 3rd party sources like HERE, crawlers on eCommerce websites with NLP, Beautiful Soup
- Extensive feature engineering using crawlers on websites & using NLP (LDA, word2vec, doc2vec) to profile apps & websites visited by users
- Location intelligence like footfalls counts & building user home locations, work locations, shopping indices, brand affinities, routes travelled
- Enriching the location intelligence through Object detection & Image Segmentation using YOLO, VGG-19 & AlexNet from satellite imagery
- Campaign optimization through DOE & Reinforcement learning (Contextual Bandits- Bayesian Thompson Sampling) to explore-exploit Bids
Product use case: Campaign conversions increased by 7% in 3months & Route analysis increased Inventory utilization by 15% for a car rental
Senior Data Scientist Media IQ (Ad Tech, Web-Display ad) [Aug’15–Oct’17]
- Launched ‘Enterprise Consulting’: team of 5 data scientists & data engineers to develop & customize analytical products & deliver insights
- Delivering advanced enterprise level insights & campaign insights to the brands for key campaign accounts
- Products/Features built for analysts to scale campaign/customer insights & provide analytical rigor through advanced modelling techniques:
- Modelling Platform: Tool built for analysts to build various predictive models, build & visualize insights & action across channels
Tech Stack: Data management layer, code in R, python, pyspark- k-means, Logistic Regression etc, visualization & UI flows
- Bot detector: built models (VW & RF) to detect fraud IP’s by time series analysis, mouse movements, click & other page level features
- Look-alike segments: Tool that maintains browser’s history (cookie ids), learns from conversion history & builds optimized target audience.
Tech Stack: HDFS, Pyspark & shell script for Feature engineering, Random Forest, Decision trees, NLP Word embeddings
- Smart Crawler: Context aware crawlers customized for a) Social networks (using NLP, Tweepy, Neo4J). b) Competitive Information
- Sentiment Classifier: Building pipeline that ingests tweets from GNIP & classifies selected tweets using Bi-directional LSTMs on Keras
Tarun Aditya, MBA, Kaggler
✆ : +91 8884534666, ✉ : tarunadityab@gmail.com. Skype: tarunaditya27
2. Senior Business Consultant Manthan Systems (Productizing BI & Predictive algos for Retail) [Dec’13–Aug’15]
- Worked as a product/analytics consultant & built various aspects of product like writing predictive model code in pyspark, python
-Algorithms productized: Churn modeling, Segmentation, Propensity modeling, Market Basket analysis, Customer lifetime value etc
-Used the product for a Retail banking client improving the 90+ day delinquency by 24% & the projected improvement in LTV was 32 %.
Program Manager, Data Sciences, Competitor monitoring Amazon (Seattle/Bangalore) [May’12–Dec’13]
Synopsis: Responsible for delivering maximum number of consumable competitive pricing inputs for all Amazon SKUs to business partners
- Customized crawling & scraping algorithms for multiple stakeholders based on pricing requirements & building SLAs & policies of pricing
- Managed improvement projects with Tech team to tune SKU mapping feeds & crawler settings- crawl frequency, scrape effectiveness etc
- Developed tool for ops team that customizes scraping from product pages via RegEx, XPATH & selenium tool
- Used scraped features from competitors like Rebates, Availability, shipping etc for internal in-stock SLA benchmarking
- Price optimization modeling & simulation to demonstrate how price elasticity, base price & competitive play affects margin
Sample Engagement: Established process to give Campaign insights & consulting multi brand CPG players to optimize sponsorship
Analytic Framework: Scoped out key deciding parameters that assist integrated campaign planning. Benchmarked campaigns size,
duration & response rates. Clustered Brand (K-means), scoped out feature vectors like- impressions, CTR, campaign length, CPC etc.
Performed Basket Affinity analysis to get category, SKU, brand associations. Used Multidimensional scaling for visualization. Interpreted
& hypothesized reasons for campaign performances & recommended grouped brands & products for integrated planning cycle
Sample Project: Establishing process to recommend competitive shipping values of products using predictive modeling
Business need: Competitors Shipping prices are key inputs for Amazon to be competitive. Web crawlers can’t extract from 54% of
competitors due to limitations (Technology issues, anonymity threat etc). Thus the need for a scalable predictive model
Analytic Framework: Chose a multivariate linear regression model (among others like Quantile regression etc) that predicts
competitors shipping prices in each category based on the prices of the competitor’s products, weight, dimensions, subcategories with
different lines fitting different segments of data. Set benchmarks of accuracy metrics (underbias, overbias, MAPE, MAD etc) through
other models like Polynomial Regression, Random forest algorithm with L1- regularization & using variable transformations
Result: Set acceptable deviations & SLAs. Automated the shipping price prediction process & scaled it to predict prices for all products
across 1500 competitors-category pairs, with an accuracy of 30% MAPE & increasing pricing accuracy of 43% of skus by 22%.
Sample Project: Improve Competitive Pricing of Newly/PreLaunched Products by modifying the Machine learning- ranking model
Business Need: The resources that provide competitive inputs to price SKUs are prioritized through Random forest that ranks the
SKUs. Existing model dint have the right features vectors for prioritizing Newly/Pre launched SKUs
Approach: Iterated with feature vectors like Age etc. Rebuilt the training data for the right classification/ranking profile i.e acceptable to
the stakeholders. Ran Pilot on limited SKU & expanded the pricing coverage (of the other success KPIs) from 2% to 48% of SKU.
MBA INTERNSHIP -4months Dell Global Analytics Project-“Redesigning Supply Chain of Enterprise Products”
- Benchmarking SCOR processes & Gap analysis. Rationalized-Skus. QA plan- Predictive Modeling of defects in new SupplyChain
WIPRO TECHNOLOGIES (JAVA & MIDDLEWARE DEVELOPER, APP SERVER ARCHITECT) [JUNE’06-MAY’10]
BFSI clients: Built & deployed SAS models in production like Credit scoring applications, Regression models, building ETL
Ecommerce Giants: Deployed SAS- Size pack Optimization & custom merchandize analytic packages for ecommerce applications like
Merchandising, Forecasting, Pricing. Customizing dashboards on Microstrategy, Cognos etc.
Academic & Scholastic
● 1st
amongst 253 teams in Accenture Bschool challenge @ IIM-A|2nd
amongst 180 teams in ACE challenge 2012 of Amazon.
● 760/800- GMAT score, 100 %il in Quant Ability & 99.36 %il in XAT(2.67 lakh test takers) & 99.82 %il in CAT(3 lakh test takers)
● Mentor on Great learning Inst reshaping the ambitions of 1283 students to leverage Data sciences to improve their work
EDUCATIONAL & SCHOLASTIC RECORD
2012 PGDIM-MBA NITIE, Mumbai (2010-12) 7.82/10 Top Scorer in Supply Chain, Econometrics, Operations Research
2006 B.Technology JNTU-Hyderabad 72.45 Top 4 & Best outgoing Student in College; AIR 29 GATE
Tarun Aditya, MBA, Kaggler
✆ : +91 8884534666, ✉ : tarunadityab@gmail.com. Skype: tarunaditya27