Inspirational social shopping experience (Inspire Tab)

● Inspire is a TikTok-like shoppable video and photo feed in the Amazon retail app, enabling customers to discover and buy products through a personalized recommendation-driven experience.

● End-to-end technical ownership of the feed stack. Defined the technical vision and led 25+ engineers across six orgs to deliver candidate-generation services, Kinesis/Spark ingestion, multi-stage ranking, and iOS/Android presentation layers while preserving <300 ms p99 end-to-end latency at 3 K TPS peak.

● Petabyte-scale hybrid pipeline. Designed a parallel batch-and-stream flow (Kinesis → S3 → Spark EMR) that cut feature-store refresh SLA from 24 h to 2 h, enabling same-day trend capture.

● Multi-level ranking platform. Built a configurable recommender model with offline candidate retrieval, SageMaker-hosted scorer that blends collaborative-filter signals with business rules for online candidate ranking and generation.

● Microservice for candidate generation. Authored a Java/Scala service with pagination, de-dup, and interest‐based filtering; runs in ECS Fargate, autoscaling to 3 K TPS with p99 <300 ms.

● Business impact. Scaled to 2 M DAU, 25 M uniques, and $32 M annual GMV while keeping p99 E2E latency < 150 ms.

>>>>> Amazon Programs and Features Personalization

● Built the platform that tailors Amazon programs (Deals, Prime, Subscribe-&-Save) in the navigation menu and homepage using ML-driven ranking.

● Full data stack ownership. Ingested billions of click & view events via Kinesis → S3 → Spark; hourly roll-ups and compaction trimmed I/O cost 60 % and cut pipeline wall-time 70 %.

● AWS-native personalization API. Java + ECS front end reads pre-computed candidate lists from DynamoDB and applies lightweight real-time rules; p99 < 75 ms at 2 K TPS while 40 % cheaper than always-on SageMaker inference.

● Impact. Personalized menu lifted CTR 50 %, drove +2 % revenue (+$35 M annualized) and cut VP escalation tickets 80 %.

● Worked on Brocade VCS cluster platform; delivered distributed control-plane services that power flat, scalable data-center fabrics.

● Designed “logical-chassis” architecture so any node can configure the entire 40-switch fabric—cutting provisioning effort ~80 %.

● .Implemented fast state-sync and transaction back-end (C++/Java), reaching sub-second convergence on topology changes.

● Published REST / NETCONF APIs with YANG models consumed by OpenStack & VMware for self-service automation..

● Rules-based API contract. Developed a standardized API and logic evaluation engine, abstracting complex conditional workflows, ensuring backward compatibility with 10+ legacy app versions and multi platform compatibility.