• Architected RL post-training framework for large-scale recommendation and LLM-style models, solving hybrid sparse–dense parameter sharding and weight synchronization challenges in distributed RL training

• Led Mixture-of-Experts (MoE) LLM models fine-tuning for multi-turn and delayed-reward training; built sandboxed environments to significantly improve experimentation speed and efficiency

• Developed agentic optimization framework for model kernel tuning and MLE workflows, automating performance optimization and accelerating model iteration

• Built end-to-end ML infrastructure (training → publishing → serving) supporting extreme sparse scaling without increasing gpu resources

• Designed flexible DAG data processing pipelines to optimize sparse weight processing flow

• Led development of core pub/sub infrastructure powering real-time notifications at Meta scale

• Resolved critical concurrency and cross-region replication issues, improving reliability and SLA

• Enhanced latency and consistency guarantees in distributed messaging infrastructure

Cloud & Hybrid Infrastructure (VMC on AWS):

• Led monolith-to-microservices decomposition, redesigning authentication (OAuth2) and async communication systems

• Built scalable hybrid cloud architecture integrating on-prem and AWS services

• Improved system availability and reduced operational overhead

Containers & Kubernetes Platform (VIC):

• Led container infrastructure for vSphere Integrated Containers (VIC)

• Designed image deduplication system (download once, reuse across VMs)

• Integrated with Kubernetes (Virtual Kubelet) to enable VM-backed pods with strong isolation

Big Data Platform:

• Tech lead for Big Data Extension, enabling Hadoop on virtualized environments

• Enhanced scheduler for data locality awareness in virtualized systems

• Filed 1 patent + 1 academic paper

Database-as-a-Service Platform:

• Built core features for Data Director (DBaaS) including HA, backup, and operational tooling