Amazon
|Software Development Engineer
Summary
Featured Merchant Algorithm, Context Aware Precompute • Export Consistency Service - Workflow for measuring consistency of Featured Offers of Exports use-case in our data stores • Designed a robust sampling strategy to obtain a list of ASINs to ensure fair representation of ASINs for workflow to run upon. • Designed and implemented the workflow to compute consistency of our featured offers in exports use-case with a latency reduced by 50 ms on an average by the use of caching over external components which vend static configuration • Deployed the workflow as a long running Fargate service on AWS ECS and set up corresponding monitors and alarms • All Buying Option Support - Project to support multiple buying options in our precompute system, which increased the accuracy of our precompute system by 1.4% • Designed an enhanced ASIN sampling strategy to enable testing of changes in order to support various buying options with a fallback mechanism to ensure representation of ASINs having buying option type with very low view traffic • Onboarded static configuration required to fetch different buying options to Amazon internal config store and developed the client required to fetch these configurations. • Enhanced the logic of accuracy computation in order to incorporate the support of different buying option types, • Segments Data Publisher - Service to publish data from S3 bucket to Amazon internal data store • Created an end to end resilient workflow from using AWS Step Functions and AWS ECS to publish segments data from S3 to our data store • Implemented multi-threading for optimal usage of the ECS Fargate computing allowing to publish data at a rate of 20000 TPS • Implemented checkpointing mechanism to ensure smart handling in case of ECS task failure and subsequent retry with Step Functions Search Relevance, Non-Default Sort • Ranking Model Evaluation Framework - Framework to evaluate various ranking models in production used to rank search results of customer queries • Engineered the framework from end to end with various AWS technologies like AWS Step Functions, AWS Lambda and AWS EMR Serverless to calculate various metrics and dump the result in a S3 bucket • Integrated the framework with a Query Replay Service via collaborating with multiple teams to automate extraction of customer queries with a high TPS and used the extracted logs for computation of various metrics required, reducing the manual effort required in evaluation by Tk person-hours • Configuration for A/B Testing of Ranking Models • Worked on configuration for A/B testing and launch of various ranking models in production, with a projected annualized impact of more than $1,000,000,000 worldwide in 2023