SRISTI SHARMA

Designed and delivered the end-to-end Video Skill Kit (VSK), enabling video playback on over 12M third-party smart TVs for Alexa's LLM-based conversational AI.

Architected an app injection layer to compensate for missing installed-app context on 3P devices, facilitating foreground app detection and content routing.

Resolved a critical playback state reporting gap by designing deterministic rules, eliminating 15% of failure rates in routing transport controls to video activities.

Instrumented device-type telemetry, exposing a 2.3s P99 latency gap in 3P regressions previously masked by Fire TV aggregation.

Owned gradual rollout and change management, strategizing targeted fixes that drove a 17% reduction in customer-perceived defect rates post-launch.

Redesigned a monolithic 1.8TB media feed validator into a distributed, parallel-processing pipeline using AWS Fargate, SQS/SNS, and MemoryDB, cutting validation runtime from 2 days to 2 hours.

Unblocked a Google Knowledge Panel launch, driving 1.1 billion incremental impressions within 60 days through pipeline optimization.

Designed and launched a new Lambda micro-service for the Track Detail Page backend, handling 10K TPM, enabling web discovery, URL routing, and GraphQL-based data aggregation.

Architected and implemented SEO metadata generation (page titles, descriptions, canonical URLs) and led interns, projected to drive 319K annual paid subscriptions.

Pioneered an automated LLM-powered meta-description engine using LangChain router chains and an AI agent, improving page discoverability by 29% through S3 & CloudFront edge-caching.

Led automation of SEO experimentation workflows across AWS Athena and SageMaker, reducing manual developer effort from 4 days to 1 day per experiment.

Implemented dynamic RTL layout support for Webplayer & Android clients, introducing bidirectional layout resolution to enable Amazon Music expansion to 5 new territories with a projected 28.7MM Total Addressable Market (TAM).

Designed a locale-gating mechanism using a custom feature flag template that decoupled string publishing from locale launch state.

Eliminated 5 hours of developer effort per release by allowing localized strings for unreleased markets to ship to mainline without activating in production.

Built a data ingestion and transformation pipeline on Azure Synapse Analytics using Python and Spark.

Repartitioned data by processing date, cutting monthly ingestion runtime from 8 hours to 2 hours.