Engineered an LLM-powered AI agent for customer support, automating the resolution of frequent inquiries by integrating custom widgets and procedural runbooks, which significantly improved response consistency and operational efficiency.

Led the migration from a monolithic to a federated GraphQL architecture to enhance system scalability and maintainability.

Designed and implemented a streamlined subgraph development framework, reducing boilerplate code and accelerating development cycles.

Optimized performance across the GraphQL gateway and all subgraphs by diagnosing and resolving critical issues such as memory leaks and thread leaks, leading to improved resource efficiency and system stability.

Developed and supported code generation tooling for frontend, Android, and iOS/Swift to ensure seamless integration with the federated GraphQL schema.

Built realtime rider pricing service for serving prices for ride-sharing product based on origin, destination, time, and business strategy with machine learning models(predicting driver payout and conversion rate) using Dropwizard framework(Java)

Built pricing plans management system for generating, deploying, deprecating pricing plans for ride-sharing product with ReactJS, Spark, Dropwizard

Supported migrating demand model for predicting conversion rate from a micro-service to Michelangelo, a Uber-wide machine learning platform

Built a micro-service to provide source of truth for driver state change and supply summary to other services including insurance, fleet management, driver earnings, etc.

Built a Go micro-service that generates driver related business metrics in realtime for market health monitoring using Kafka and Cassandra, supports 200k TPS at peak

Built a data backfill pipeline for blocking driver/rider if you gave/received 1-star rating to/from this person with Spark, and Piper(a job-scheduling service)

Ingesting global drivers' location update and generating events about drivers entering/existing/dwelling in geofences

Offer Service

Offer Service Pre-computation Reactors

Remote Catalog Service

Catalog Responder APIs