I currently work as a technical leader in Apple Cloud Elastic Disk team at Apple. Our service provides highly available and durable network block devices as a service at exabyte scale. As a technical leader, my daily responsibilities include collaborating with customers, designing and developing new software systems, mentoring senior engineers, and providing technical guidance to senior leadership.Some notable projects I’ve worked on in this team include:- Designed and developed a SQLite-based metadata store for service metadata, eliminating the need for persistent storage. Built a framework to integrate data collectors from various metadata stores, enabling background processing tasks and monitoring critical service performance.- Designed and developed procedural macros in Rust for SQLite, automating client object generation for each SQLite table and reducing developer time.- Designed and developed a modular background task processing service, utilizing the Reactor pattern and Tokio for asynchronous task processing. Enabled developers to create state machines for various tasks, including garbage collection, data audits, repairs, backup, and restore.- Designed and developed an authentication and authorization framework, supporting mTLS authentication and allowing developers to write complex policies based on certificate fields and extensions.- Designed and developed distributed tracing for our services, enabling accurate incident debugging.- Designed and developed an integration testing platform based on Minikube, leading to significant cloud savings and a 10-fold reduction in AWS Elastic Compute bill.

As a founding team member of Amazon Kendra, I helped build a multi-tenant, enterprise-grade semantic search engine that enables organizations to search across diverse data sources, from production databases to full content management systems. Kendra empowers users to integrate data seamlessly, allowing efficient, secure knowledge discovery across their data silos.As the tech lead for Kendra’s connector component, I focused on simplifying data ingestion from various content management systems, such as Microsoft PowerPoint, Confluence, Jira, ServiceNow, GitHub, Dropbox, and Slack. These connectors save customers the time, cost, and operational burden of custom integrations, allowing data to be ingested into Kendra with minimal setup.Key contributions:1. Connector State Machine: I designed a flexible workflow engine to support data ingestion for various connectors. This engine enables secure data transmission within AWS by allowing connectors to interface with customers’ Virtual Private Clouds (VPCs).2. Connector SDK Development: I replaced third-party software with a custom Connector SDK, reducing operational costs and simplifying maintenance. The SDK offers a generalized interface, allowing developers to create connectors more quickly. This platform also enables contributions from vendors, resulting in the development of over 40 connectors that are now used by hundreds of customers to power internal search engines and chatbot applications in Amazon Kendra and Amazon Q.3. Identity Crawler: I introduced an Identity Crawler feature in the SDK to capture user groups and permissions. This enhancement supports the creation of access-controlled search applications, enabling enterprise customers to enforce security policies on search results.Through these initiatives, my work on Amazon Kendra contributed to a highly scalable, secure search solution, helping customers unlock insights from their data with ease and precision

Senior Software Engineer on the Customer Reviews team, where I contributed to building and optimizing the core data platform for Amazon’s customer reviews and ratings. My work centered on mining, inferring, and simplifying data to enhance the consumer experience with meaningful insights. Key projects included: 1. Led a significant migration of our data platform from Amazon’s legacy infrastructure to AWS, resulting in substantial cost savings by deprecating Oracle databases and scaling effortlessly for peak traffic during Black Friday and Prime Day sales. 2. Supported the launch of Reviews Ranking Algorithm 2.0, leveraging Redshift and Amazon Kinesis to process and rank customer reviews, enhancing the platform’s accuracy and responsiveness. 3. Developed a multi-dimensional ratings aggregator, transitioning from Redshift to an EMR-based solution for substantial compute savings—up to 25x. This aggregator processed trillions of data points across millions of products to deliver precise, user-friendly ratings summaries by region, interest, and more.

During my tenure in the customer support team at Zenefits, our primary objective was to promptly address customer inquiries. As part of this team, I spearheaded several projects, including:- Development of an AI-Driven Article Recommendation System: I designed and implemented an independent microservice based on LSA (latent semantic analysis) machine learning algorithm. This article recommendation system was a pivotal component of the customer support page. By directly guiding customers to relevant articles, it effectively reduced the ticket queue by 75%.- Technical Lead for Customer Support Page: I oversaw the entire workflow for customers landing on the contact us page. The form employed a decision tree approach to gather customer information, enabling efficient routing to appropriate options.- Integration of Customer Support Tickets into Salesforce CRM: I spearheaded the integration of customer support tickets into Salesforce CRM, streamlining the communication process and enhancing overall operational efficiency.

Scalable Machine Intelligence

Machine Learning Team

At Qualcomm, I was a member of the multimedia audio team, responsible for developing audio firmware software for the Qualcomm Snapdragon processor family. As part of the audio team, we designed device drivers and optimized algorithms for emerging chipsets. Some notable projects I contributed to include:- Porting and optimizing the Adaptive Acoustic Noise Cancellation algorithm on Qualcomm’s Hexagon V5 DSP.- Implementing the Feedback Speaker Protection Algorithm on Qualcomm’s DSP.- Collaborating on the audio startup of the MSM8916 mobile processor.- Developing a highly flexible and dynamic thread launch service for the Audio Front End (AFE).- Conducting extensive on-target profiling of the Audio Front End Manager to obtain precise MIPS (million Instructions Per Second) information.

Tegra SoC Software Development

Developed a prototype in C for Viterbi Algorithm