Sukjoon Oh

My research focuses on efficient vector search and data-intensive systems. I study how caching and workload locality can reduce data movement across memory and storage hierarchies, with ongoing work extending these ideas to GPU-based vector search.

I studied RDMA-based replication and distributed systems, focusing on reducing communication overhead in state machine replication. I also worked on scalable RDMA replication systems and integrated research prototypes into Redis and Memcached.

I built my foundation in operating systems, systems programming, and computer systems, and gained early research experience as an undergraduate research intern at Corelab.

Seoul, Korea Cumulative GPA: 4.26/4.50

I continued collaborative research on approximate caching for vector search. The work resulted in Aker, a cache integrated into PostgreSQL pgvector that improves both search efficiency and accuracy, and was published in PVLDB 2026.

I worked on optimization for vector search, studying the I/O behavior and read amplification of approximate nearest neighbor search. This work explored result reuse and approximate caching as a way to bypass expensive index traversals.

Seoul, Korea

I developed operational software in C++ while serving in the Republic of Korea Air Force. Working with existing system constraints and validation requirements gave me early experience in building software where reliability and compatibility mattered as much as functionality.

Aker is an approximate result cache for disk-based vector search that adapts reuse decisions to individual queries and efficiently refreshes cached results under updates. Integrated into pgvector, Aker improves recall by up to 64 percentage points and QPS by up to 3.2× while using 0.6× the memory of PostgreSQL shared buffers.