Designed and built a new impression logging service to record advertisement views, landing page hits, and application clicks for Amazon Payment products which receive more than 50K TPS worldwide. Created separate data streams and schemas for each client which reduced the blast radius to a client level and decreased onboarding time for new services from 2 months to 2 weeks. Leveraged AWS Kinesis and Batch to deliver impression data to the Data Engineering team in near real time for processing which solved a long term scaling bottleneck. Discovered and fixed an extraneous impression logging call in the current system costing the company 250K annually.
Devised a new dynamic linking architecture plan and collaborated with stakeholders from three partner teams to determine the long term vision for how payment product application links would be configured, stored, and shared in the Payments Org. Remodeled the existing configuration pattern to reduce the amount of fields marketers have to configure by > 80%, resulting in a decrease in time needed to launch new campaigns and the probability of a misconfiguration. Expanded the current data model to support flexible schemas allowing Product teams to configure more complex campaign types that will unblock and expedite the launches of three campaigns expected to generate 130+ million dollars annually.
Engineered key components and helped launch a service that optimizes the incentive amount advertised to acquire a customer. Designed and built the data ingestion and incentive assignment tools used by the model training pipeline. Optimized the preprocessing step of the incentive assignments pipeline to decrease the steps runtime from 10 hours to 30 minutes which reduced monthly hardware costs from $10K to $2K. Reduced the cost per acquisition by 20% and increased acquisitions by 15% for 10+ clients through this launch.
