Ekko: A Large-Scale Deep Learning Recommender System with Low-Latency Model Update
Contribution: model-aware dissemination + scheduling + SLO-aware placement to keep update latency in seconds at extreme scale.
At a glance
- ML systems engineer focused on production constraints: freshness, latency, and cost.
- Co-first author of Ekko (OSDI ’22): second-level model freshness at extreme scale.
- Industry: Senior SDE at Tencent (WeChat); open-source: LLVM contributor.
Research interests
- Model freshness at scale (update pipelines as a first-class systems problem).
- Efficient serving (latency SLOs, memory bandwidth, infra cost).
- Low-cost data/feature pipelines (safe execution + locality-aware design).
Selected publications
Dynamic Barycenter Averaging Kernel in RBF Networks for Time Series Classification
Contribution: kernel design for time-series classification in RBF networks (early research work).
Experience snapshot
- Tencent (WeChat) — Senior SDE, ML systems for large-scale production.
- LLVM — contributor (commit access), performance/infra improvements.
Service
- Reviewer: CVPR 2025; ICLR 2025 Workshop; NeurIPS 2025 Workshop.