Partnered with strategic customers to understand their goals, translating ambiguous business needs into clear technical plans, and aligning cross-functional teams to deliver roadmap-shaping POC initiatives.

Secured new contracts by leading the delivery of production-ready POCs for high-profile clients, defining critical user events and flows in their agentic AI apps and uncovering actionable insights that resolved core product pain points.

Influenced direction of product based on customer needs by championing semantic extraction from AI agents’ chats.

Developed a custom Rust-based real-time platform designed to support app analytics that processes and aggregates hundreds of millions of active sessions for downstream services.

Reduced response time and memory usage by 80% across multiple company-wide services by implementing a Redis cache on a real-time configuration microservice.

Architected and owned a highly available real-time configuration-pulling service, significantly decreasing infra costs while ensuring zero-defect delivery and increasing the stability and observability in Python.

Designed LLM-driven workflows to automate the development of domain-specific analytics metrics by prompt engineering Claude Code, reducing development time from weeks to hours.

Building a customer-facing RAG-powered Al agent that answers product questions via graph database queries and captures sales leads via Salesforce integration.

Owned a video analytics platform that processes 2 GB/s, ensuring 99.99% up-time and high throughput across hundreds of business customers including major streaming services like Hulu, Sling, and NBC.

Drove end-to-end release process for a distributed real-time service with coordination of multi-geo/function engineering teams, and ensuring rigorous testing, monitoring, and risk mitigation for feature, stability, and performance updates.

Designed and deployed robust ETL pipelines utilizing Spark and Kafka, processing and transforming large-scale data for real-time analytics, leading to a 30% decrease in data processing latency.

• Operate and maintain NRTG (Near Real-Time Galaxy), a system that allows creating fast, systematic, and scalable Flink application built on top of our custom Dataflow API at Pinterest in 2019, with an impact of over 70 client’s day-to-day work and powered by millions of budgets.

• Completed an end-to-end upgrade of Flink jobs’ metrics exposure, including implementation of local metrics test-agent, customization of metrics with tags, reporting metrics in OpenTSDB with retry mechanism, etc., that brings a positive impact to hundreds of hosts.

• Optimized Flink-based system (NRTG) with down sample function to improve performance by reducing certain sink function’s QPS from millions to hundreds.

• Independently designed and implemented the at-least-once semantics of async operation for Flink jobs, which helped improve streaming jobs with a relatively stronger consistency along with handling RPC failures of finagle services and Kafka data loss.

• Lead and executed an end-to-end upgrade on monitorization of workflow’s health and vitals that will save 50% time on onboarding alerts and finding the root cause of issue and impact to downstream services.

• Guided clients from different team with customized support like onboarding streaming jobs, performance improvement, and solutions to difficulties and challenges in day-to-day operations.

• Developed customized Flink sink for key-value storage with widely differing functionalities to adapt state-of-art tech stack.

• Take part in team's on-call rotations entailed supporting and root causing incidents across new and legacy technologies and prevent server incident from happening.

• Supported NRTG adapt new sinks from different teams by performing evaluation and integration tests with cross-team collaboration.

Designed and built a tool to fetch the messages and status metrics information of Kafka clusters.

Implemented an UI to visualize the Kafka process status using React.

Recommended by the department to be a one-on-one tutor for students in computer science and math classes.

Helped students with their questions during the weekly lab session, and graded homework.

Assisted students visiting the help room and provided help on homework and lab.