Led Redis Infrastructure Architecture and Uber Java Client Integration:

• Designed and developed a robust Uber Java client wrapper for Redis, abstracting interactions over Jedis and Lettuce libraries to support a wide range of application use cases.

• Contributed key enhancements to the Lettuce open-source library, focusing on connection leak prevention and topology refresh mechanisms, improving Redis cluster reliability at scale.

• Spearheaded the migration from Twemproxy with standalone Redis to a high-availability Redis Cluster setup, reducing latency and improving fault tolerance across Uber’s distributed services.

Architected Multi-Region Resilience for Uber’s Workflow Orchestration Platform (Piper – Uber’s internal version of Airflow):

• Designed and implemented multi-region failover capability in the Piper system, enhancing disaster recovery and ensuring workflow execution continuity under regional outages.

• Collaborated with infrastructure and platform teams to deliver high-availability architecture aligned with Uber’s global operational requirements.

Architected and led the standardization of DoorDash’s in-house real-time Customer Data Platform (CDP):

• Unified the event processing ecosystem by integrating Confluent Schema Registry, Kafka REST Proxy, Flink, Kafka, and Snowflake into a cohesive, scalable architecture.

• Designed a dual-format schema strategy using Avro and Protobuf to enforce data contracts across internal services.

• Built a modular pipeline architecture enabling per-event stream processing and isolation, ensuring clean delivery into Snowflake with strong schema guarantees.

• Leveraged Kafka REST Proxy to enable JSON-based external event producers without compromising schema compliance.

Designed and implemented a real-time event forwarding service from DoorDash to external partners:

• Guaranteed at-least-once delivery semantics using Flink checkpointing and Kafka offset management.

• Introduced a Dead Letter Queue (DLQ) strategy to gracefully handle external system outages and improve observability.

Led the migration of 50% of traffic from a 3rd-party CDP to the in-house real-time pipeline:

• Reduced operational cost and increased data governance and system reliability through internally owned infrastructure.

Enhanced observability and developer productivity for real-time data pipelines:

• Extended Apache Flink SQL with custom tooling to support Protobuf format inspection via Confluent Schema Registry.

• Developed an automated schema evolution handler for Avro, integrating Schema Registry, Snowflake SDK, and Cadence workflows to ensure backward and forward compatibility in production pipelines.

Natural Language Process platform

Medical Knowledge Graph

Medical AI based chatbot

In Medical AI Lab, lead Engineer team, build medical AI based platform/ product

Neo4j/MySQL based Medical Knowledge Graph

• Neo4j store the medical entities and relationships

• MySQL for entity normalization

• REST API for internal data services

• Dynamic construct Knowledge Graph by merging different sources, based on configuration

Medical Natural Language Process

• Entity data dictionary/word segmentation based on Knowledge Graph

• Multi-dimensions information retrieval for major entity

• Various medical specific parser: prescription parser, medical exam index parser

AI based medical chatbot

• Parse patient main symptoms with Natural Language Processing

• With user profile/AI model, generate most distinguishable question for user to answer

• After multiple rounds, generate disease prediction report

Big Data Lab: Medical AI based chatbot

Disaster Recovery for LinkedIn main production ETL pipeline: Lumos

Dataset Configuration Management in Apache Gobblin

Regression testing framework for LinkedIn version of Distcp: Uber Distcp