Hi, I'm Jaesol, a research engineer in Seoul working on agent tooling

I'm Jaesol Shin, a research engineer at WeDataLab in Seoul. I work across AI infrastructure, databases, and agentic systems, with a focus on making AI observable, reliable, and practical to deploy. My work spans RAG and knowledge retrieval, PostgreSQL extensions, GPU-accelerated search, LLM observability, and agent runtimes.

I contribute to open-source projects including LightRAG, and build tools such as pg_cuvs (GPU vector search for PostgreSQL), cctrace (self-hosted observability for Claude Code and Codex), journey-qa (agent-driven acceptance testing), and xsm (cross-session messaging). I also maintain a research notebook, Latent Space, where I write about agent architectures, retrieval benchmarks, inference optimization, and the engineering trade-offs behind AI systems.

Lately, I've been thinking about how software should change when general-purpose agents become the workers. Rather than embedding a separate AI assistant into every product, I'm interested in exposing specialized capabilities as reusable tools and building shared agent runtimes that can operate across products and environments, including on-premises and air-gapped systems. My background in sociology also informs my longer-term interest in multi-agent coordination and collective behavior.

What I'd love from this community: exchanging ideas with founders and engineers building agent-native products, particularly around tool interfaces, runtime architecture, observability, and how these systems evolve when deployed in real organizations.