Building GitHub-native agent workflows, MCP tool layers, and data infrastructure for testable AI-assisted delivery.
- Gira: GitHub-native control plane for issue -> branch -> PR -> checks -> completion evidence
- MCP/tooling: typed tool interfaces, versioned specs, fixtures, and agent-ready execution paths
- Datapan: private data platform workbench for Korean public/domain data, research corpora, and agent-ready data workflows
- AI-assisted delivery: tests, lint, CI, review loops, and operational guardrails around agent work
Datapan is my private data platform workbench. The public signal is the range and structure of the collection axes, not only raw volume.
- Public and administrative data workflows
- Legal, assembly, policy, and regulated-domain data
- Real estate, geospatial, routing, weather, and app-supporting feature marts
- Market, macro, finance, and investment research data
- Research-paper metadata and literature corpora, including arXiv, OpenAlex, and HuggingFace Papers style pipelines
- Raw ingestion, canonical schemas, marts, coverage checks, and data lineage for agent-ready downstream use
Recent upstream work includes OpenHands Software Agent SDK, OpenHands, and LiteLLM:
- OpenHands/software-agent-sdk#2936: FileStore fallback warning behavior
- OpenHands/software-agent-sdk#3252: custom FileStore injection path
- OpenHands/OpenHands#14028: Gemini base URL integration issue
- BerriAI/litellm#26979: Gemini custom api_base duplicate path issue
- statpan.com for technical notes, project notes, and longer-form engineering context


