about
I build production LLM/RAG systems and browser 3D engines.
AI Engineer at Redbrick in Seoul. I ship end-to-end AI pipelines from prototype to production — agentic workflows, RAG over large document corpora, semantic caches, 3D game engines. Grounded in an HCI research foundation: 10 peer-reviewed papers, 6 patents.
Experience
2021 — now · AI / 3D / research, always in production.
AI Engineer · Redbrick
- End-to-end AI game creation system (intent parsing, asset generation, scene composition, logic synthesis) — centerpiece of a $1.2M Korean government R&D grant.
- Production RAG pipelines over 15K+ documents in ChromaDB with hybrid retrieval, query rewriting and bge-reranker — relevance +35%, hallucination 18% → 4%, latency 3.2s → 0.8s.
- Multi-step agentic workflows (ReACT, CoT) orchestrating LLM calls across asset generation, physics config, and script authoring — up to 8 dependent tool calls per request.
- Semantic caching layer (embedding similarity) — ~40% fewer LLM calls, ~$2K/month saved.
- Eval & safety : prompt regression harnesses, output guardrails, structured validation, policy-compliant generation.
3D Frontend Engineer · Redbrick
- Browser 3D game engine (Three.js, WebGL, TypeScript) powering 54M+ game plays and 10M+ user sign-ups.
- Codebase modernization of 50K+ lines from Webpack → Vite, upgraded rendering pipeline, swapped physics engine — build time −60%, bundle −35%, hot-reload 8s → 1s.
- Modular packages across 3 teams; integration tests for scene graph and asset loading.
- Onboarding redesign via session replay + interviews — D7 activation +25%, D30 retention +15%.
Embedded Systems & HCI Researcher · DECS Lab, UNIST
- IoT firmware in C for ARM Cortex-M MCUs, real-time smart sensor arrays (50+ devices, sub-ms latency).
- 10 peer-reviewed papers + 6 patents . First author at IEEE RO-MAN 2023, ICROS 2022, HCI Korea 2022.
Education
Design × CS, HCI/HRI focus.
M.S. in Design (Human-Computer Interaction) Feb 2023
UNIST, South Korea
Thesis: Human perception of social-robot face and color expression using computational emotion models.
B.S. in Computer Science & Industrial Design 2021
UNIST, South Korea
Stack
What I reach for, in order of depth.
Languages TypeScriptPythonJavaScriptC / C++
AI / LLM LangChainLlamaIndexVercel AI SDKChromaDBOpenAIClaudeEmbeddingsRAGAgentsEvals
Frontend & 3D ReactNext.jsAstroThree.jsWebGLViteNode.js
Infrastructure DockerAWSRedisCI/CDGit
Research HCIHRIComputer VisionEmbedded CARM Cortex-M
Certifications
Recent focus: RAG, agents, vector DBs.
- → Advanced RAG with Vector Databases · IBM · 2025
- → Foundation — LangChain (Python) · LangChain · 2025
- → Gen AI Agents · Google Cloud · 2025
- → Build RAG Applications · IBM · 2025
- → Three.js Journey — Completion · 2023
- → Advanced React · Meta · 2023
Why
I'm not the person who'll build your fourteenth demo. I'm the person who makes demo #1 a product — eval harnesses, cost dials, guardrails, the boring parts that decide whether an AI system ships or stalls.
Want the fast version? Talk to td-agent.