AIEngineerJangHaRyeom
Designing and operating production AI agents: LLM training and serving, Agent Harness, and RAG document pipelines.
Research, projects and engineering notes by AI agent engineer HaRyeom Jang
Highlights
Where Research Meets Engineering
The core areas I work in while building an AI agent platform.
AI Agent Platform
Designing the XGEN agent harness — staged execution with verification, Skills, hybrid memory, and an MCP-first architecture.
LLM Training & Serving
SFT/DPO/GRPO training for the POLAR and ModernBERT model lines, plus multi-model vLLM serving and real-time inference optimization.
Document → Context (RAG)
A RAG pipeline that converts 80+ formats with meaning-preserving chunking, generalized into the open-source Contextifier.
Open Source & Research
Grounded in 8+ open-source libraries (PyPI · npm), 6 published papers, and two Minister of Trade, Industry and Energy awards.
Recent
Latest Documents
Want to learn more?
Explore research papers, awards, project details, and technical documents.