Senior Software Developer in Seattle, WA, United States

Amazon Cash is an application that allows Amazon customers to add cash to their Amazon Balance. A customer is assigned a unique barcode to take to participating retail stores to add cash. The customer's account is credited in retail time, and the customer can use their balance to purchase millions of items at Amazon.

I was the lead developer of the project. I designed the client-facing APIs and internal systems to process transactions from retail stores. I also led a group of eight engineers to develop the APIs and internal systems. I implemented a few critical APIs, such as credit and refund of balance to/from customers' accounts and service to notify customers when their account is successfully credited. I designed the barcode generation and validation algorithm.

I was responsible for creating the API specification and integration document for clients. I led the technical discussions with partners to onboard the APIs to their system. I was a point of contact during their integration and testing. I also conducted product beta testing and visited retail stores in a few states to test the product.

Project launch media reference: https://money.cnn.com/2017/04/03/technology/amazon-cash/index.html

SNAP is a federal assistance program for eligible low-income individuals and families to purchase certain foods in authorized stores. SNAP funds can be accessed through an EBT card. Historically, EBT cards were only supported at physical retail stores, but in 2017, Amazon got approval to pilot the SNAP program, and my team was responsible for the development work.

I co-led the design of changes required in Amazon Ordering and Fulfillment pipelines to support EBT cards and led a group of seven engineers across three teams to implement the changes. I was also responsible for updating the payment execution engine, which processes all payment transactions in Amazon, to support a product-restricted payment method. My work laid the foundation to help other product-restricted payment methods globally, such as Health Benefit Cards in the US, Product and Food Vouchers in the US and Mexico, and so on.

SNAP program has become an enormous success enabling millions of low-income customers in the US to purchase foods online via Amazon Fresh and Whole Foods during the pandemic.

As an engineer in the Amazon Gift Certificates team, I co-led the design of a highly scalable balance tracking system backed by AWS DynamoDB and a NoSQL database to replace a legacy balance tracking system backed by a relational database. The legacy system had performance issues due to its database, which was becoming a scaling bottleneck.

I led refactoring of the Gift Certificates services to migrate them off the legacy system to the new system. As a part of this effort, I designed and implemented a validation framework to prevent the processing of migrated account transactions on the legacy system. I also implemented a continuous migration strategy that migrated a small set of randomly selected customers to the new system every day.

In the project's final phase, I co-led an effort to migrate millions of customer accounts and their transactions from the legacy system to the new system. I co-authored a library that atomically migrated a customer account and corresponding transactions to the new system. We successfully migrated hundreds of millions of customers globally without any downtime.

As a senior engineer in the Geo-sciences and Maps team at Amazon logistics, I was responsible for designing and implementing an orchestration framework to automate model training and inference. I used AWS Step Functions as a workflow engine and AWS Lambda to proxy a job in each workflow step. For the actual computation in each job, I used AWS Batch and AWS SageMaker. The framework also provided plugins to extract training and inference data directly from Amazon S3 and Glue. I also developed an internal tool to create and publish workflow definitions in AWS Step Functions.

The framework laid the groundwork for developing multiple machine learning models. The framework supported multi-tenancy and could be used to run workflows in multiple AWS accounts.