Experience

University of Toronto · Computer Engineering Group ↗

Research Intern · Flink autoscaling · Stateful operators

Advised by Prof. Ashvin Goel

Apache Flink is a distributed stream-processing system whose Kubernetes autoscaler uses DS2’s performance model and operator-rate estimates to determine parallelism. Building on Justin’s hybrid CPU/memory autoscaling approach for stateful operators and its trial-and-error reconfiguration challenge, we are building a model that directly derives efficient CPU and memory scaling rules at runtime while preserving resource efficiency.

NTU Secure System Lab ↗

Undergraduate Researcher · SEV-SNP · vsock · eBPF

Advised by Prof. Shih-Wei Li

1. Vsock-net: Making Paravirtualized Network I/Os for Linux-based Confidential VMs Safe and Fast (Preprint, 2026)

Removed the large, attack-surface-heavy Linux network stack from confidential VMs’ Trusted Computing Base (TCB) by delegating network traffic to the host stack. Despite the performance overhead typically introduced by delegation, optimized virtio-vsock communication and leveraged eBPF sockmaps for zero-copy packet redirection in the host kernel, achieving 12.59× higher throughput and up to 10× faster Nginx performance under TLS encryption.

NCKU Office of International Affairs

Backend Software Engineer · Agentic systems · Redis

• Built the backend for an agentic assistant serving 500+ students, with a custom MCP server for document retrieval.
• Optimized frequent-query handling with FAISS-based semantic caching, improving end-to-end throughput by 3×.
• Designed Redis-coordinated asynchronous worker pipelines to scale concurrent agent workflows.