AdTech - Sponsored Products

– Lead design, implementation & integration of foundational system for Dynamic Bidding, AutoPilot, TROAS Bidding etc., for Sponsored Products(US Patent application in progress), contributing to potential increase of 5-8% (around 120M USD) in annual Ad revenue

– Lead design, implementation & integration of Darwin, new retrieval service with 1B req/day (10K QPS) using event driven Spring Webflux framework (resource utilization halved saving 200k USD / year) to decouple retrieval logic from ranking, scoring and filtering functionality of Abram, Retrieval and Ranking service.

– Lead design, implementation & integration of Brand Conquesting for Sponsored Products, to allow advertisers to conquest competitor keyword search.

– Lead design, implementation & integration of Cold start for Sponsored Products, contributing to 50% boost in under-performing adItem impressions with less than 0.5% drop in Ad revenue.

– Designed, implemented other features like Solr A/B test tuning, Feature level caffeine cache refresh

– Tech Stack : Scala, Java, GCP, Kubernetes, Docker, Azure, WCNP, Splunk, Mleap, Druid, Feast.dev

(MLOps)

Universal Image Feature Catalog (UIFC) - ML Platform (Classification & Policy Platform)

– Lead design, implementation, onboarding & integration of AlexaAI team’s ShopTheLook, CPP team’s

ProductType classification use cases to UIFC ML platform using Protobuf schemas.

– Lead implementation for Precache API in ImageRetriever to cache downloaded image & metadata to

reduce MSA (Tier-1 service) API calls by 75%

– Lead operational improvements like saving 600K+ USD/yr (88% reduction) in DDB storage costs using

compression, serialization and trimming etc.,

– Designed, re-architected, and implementated DSL schema (protobuf) for optimizing feature storage &

SerDes times for Facade, Feature store & Precompute services (part of UIFC ML Platform)

– UIFC ML platform productionizes ML models (250+ so far) provided by applied scientists, runs ML

models and stores generated features (for 20B+ catalog images), serving 100K+ TPS traffic

– Tech Stack : Java, AWS (SageMaker, Batch, DynamoDB, CDK, Lambda, SNS/SQS, Step Functions, API Gateway, CloudAuth, Hydra, Kinesis, ECS/Fargate), Protobuf, Google Guava/Guice, Micro ser-

vices

Associate Data Services (Affiliate Marketing)

– Worked on migrating data of around 3 Billion records of Associates data from RDS to DynamoDB

– Acted as Lead SDE for our team (with 24 clients) during 2021 Peak and carried out scaling, disaster

recovery planning, chaos testing etc., to ensure a successful peak

– Designed & implemented asynchronous workflow (Firehose) using SNS & SQS messages for supporting

Global StoreId project & GDPR compliance efforts.

– Tech Stack : Java, Micro services, AWS (RDS, DynamoDB, Glue, Athena, Data Pipeline, Lambda,

SNS/SQS, Hydra), ANDES data lake

Ad Experience Cloud - Search & Display Advertising

Adobe Media Optimizer

– Owned, re-architected and optimized execute module (that applies keyword bids to Search engine) of

Adobe Media Optimizer (Search Ads) team and implemented YDN (Yahoo Ads) execute.

– Designed & implemented hybrid optimizer that combines Adobe Adcloud bid optimization and pub-

lisher platform (Google Ads) advertiser goal optimization.

– Improved process syncing keywords, clicks from Google Ads, Bing Ads to run 50% faster by imple-

menting new hashing mechanism. Resolved bottlenecks in adding new SE entities.

– Helped gain 1M+ USD revenue to AMO in Jan 2019 by enabling Google optimized campaigns and

increasing Ad spend on Google managed campaigns.

– Optimized a critical module ’materialize ads’ to run 10x faster, it handles millions of keywords daily

to generate search-engine agnostic bid units.

– Implemented APIs for assigning campaigns, adgroups, keywords to portfolios from new AngularUI.

– Designed & implemented ’Events’, that automates the budget allocation of campaigns and keywords

during the peak holiday season based on bid history.

– Tech Stack: Python, Flask, PostgreSQL, Kafka, AirFlow, Splunk, Prometheus

– Implemented modules for Tizen TV mobile web browser being developed from scratch using WebKit-efl

& Chromium-efl, EFL graphics libraries and C++.

– Contributed to KPI improvement of Canvas, CSS3 and JS modules of Chromium-efl & WebKit-efl.

– Fixed several issues related to WebKit-efl, Chromium-efl rendering engines for TV and mobile.

– Tech stack: C++, Webkit, Chromium, EFL, Java, MySQL, Embedded-systems