Meta Superintelligence Labs

• KV Cache systems lead (Inference group)

• Research Enablement core engineer

• 10k+ large-scale agentic synthetic data generation

• RL Post-training Infra (sampler/rollout)

• Performance engineering - KV Cache, P/D Disagg, Scheduling, Memory

• Production (Online) Inference serving runtime

Authored KV Cache Offloading adopted by Meta on vLLM v1 (and RL’s).

• SOTA performance. Zero data corruption.

• Supports Hybrid KV, multiple scheduling optimizations; various kinds of attention types, kernel backends, hardwares, CudaGraph modes, distributed setups.

YouTube Shorts Ads

Pipelines of system-generated Ads creatives

Monetization - experiments, formats; Fixer

Intern (International) at Data Infra Shanghai - Rheos Event Streaming.

COVID gap year