Building ontology-grounded knowledge graphs and ontology search algorithm.
Agent env researcher, such as multi-agent harnesses/forge/loop that run on them.
Ontology & knowledge graphs: building backend-agnostic ontology / KG toolkits end to end — document and tabular extraction, entity resolution, predicate governance, dedup, is-a hierarchy induction, quality scoring — then retrieving over them with one-shot GraphRAG (vector + graph-label + class-enumeration fusion, MMR diversity, adaptive top-k instead of fixed-k). Zero-infra by default, any SPARQL 1.1 store when you outgrow that.
Multi-agent systems: agent cognition (persona, emotion, memory, theory-of-mind), harness execution engines, and forge engineering — agents that rewrite their own harness under benchmark-gated control.
Open source and research: js-ontology-build / js-omnifuse / xgen-harness / Agethos on PyPI, multi-award academic publications.
- Taxonomy induction, no LLM call: Korean Hearst-pattern hypernym discovery + head-noun compound decomposition, straight out of raw documents and tables.
- FK-aware table-to-ontology: star-schema fact/dimension split; FK direction resolved from the actual primary key, not naive value-overlap.
- Adaptive retrieval: dynamic score-cutoff (not fixed top-k) + MMR diversity, so minority/contradicting evidence survives.
- Backend-agnostic: zero-dep in-memory graph by default, drop-in to any SPARQL 1.1 store.
- Agent cognition (Agethos): OCEAN personality + PAD emotion driving actual behavior, Hebbian memory, vicarious learning.
- Multi-agent debate (agent-colosseum): agents debate/red-team/peer-review each other, benchmarked against single-agent baselines.
- Harness engineering: a declared
HarnessConfigcompiles to a 10-stage pipeline, workflows compile to installable MCP wheels. - Forge engineering: agents that rewrite their own harness config under versioned, benchmark-gated control.
| Project | Description | Links |
|---|---|---|
| js-ontology-build | Backend-agnostic ontology / knowledge-graph toolkit — parse documents or tables, build a clean KG (entity resolution, predicate governance, dedup, is-a hierarchy induction, quality scoring, community detection), then search it with one-shot GraphRAG. Zero infra by default (pure-Python in-memory), loads into any SPARQL 1.1 store. Published on PyPI as xgen-ontology. Source-available, all rights reserved |
  |
| js-omnifuse | Backend-agnostic one-shot GraphRAG — fuses vector + graph (label / class enumeration / relation) seeds with MMR diversity into a single synthesis; zero-infra (in-memory BM25) or any SPARQL/Fuseki. Plus Vault, an omnifuse-native memory (fuse / surface). The search half of js-ontology-build, extracted standalone. Published on PyPI as xgen-omnifuse. Source-available, all rights reserved |
 |
| xgen-harness | Declarative LLM agent execution engine (harness engineering) — declare a HarnessConfig, get a 10-stage pipeline. Multi-provider, capability-based tool matching, compile workflows to installable MCP wheels |
 |
| JINXUS | Hyper-personalized multi-agent AI assistant — 28 agents, virtual pixel office, 225 tools, autonomous execution |    |
| Agethos | A brain for AI agents — OCEAN personality, PAD emotion, memory stream, Hebbian learning, vicarious learning, cross-platform export |  |
| Year | Award | Conference |
|---|---|---|
| 2025 | Excellence Paper Award | Korea Society of Electronic Commerce & Smart Media Society |
| 2024 | Excellence Paper Award | Korea Society of IT Services |
| 2024 | Best Paper Award | Korea Intelligent Information Systems Society |
| 2024 | Excellence Paper Award | Korean Academy of Management |
| 2023 | Excellence Paper Award | Korea Society of Information Systems |
| 2023 | Best Paper -- Honorable Mention | KHU Big Data Graduate Student Conference |
| 2023 | Grand Prize | Korea Knowledge Management Society -- Idea Competition |
| Education | M.S. Big Data Analytics — Kyung Hee University / B.S. Statistics — Jeonbuk National University |
| Experience | Plateer — AI/LLM Engineer (current); ontology engineering & GraphRAG (knowledge-graph build pipelines, backend-agnostic retrieval) and agent harness execution engines Shaveron — FDE (Forward Deployed Engineer) LG CNS SINGLEX Strategy/Operations Team @ LG Science Park — RAG Development |
| Interests | Multi-Agent AI Systems, Agent Cognition, RAG, NLP, Time Series Forecasting |
Languages
AI & Data
Backend & Frontend
DevOps & Infra
