I build AI systems that hold up in production.
I connect RAG, document processing, vector databases, and MCP across backend and infrastructure.
- LIVE
- 3
- 5
- 2
Reproduce the problem.
Redesign the system boundaries.
Verify the full production path.
Selected systems tested in production
Work where I reproduced the problem, redesigned the structure, and made the result dependable for real users.
-
Generative AI (K-Bot) Per-Type Parsing & Chunking Pipeline
A preprocessing system that lifts RAG quality by routing parsing, chunking, and embedding per document type
Designed a RAG document-processing structure that separates preprocessing paths by document type and usage context Backend · AI/RAG · 2025.11 ~ 2026.02 -
Generative AI (K-Bot) Metadata-Based Document Source Tracking
An end-to-end grounding system that traces inline LLM citations back to chunk metadata and the exact source
Linked answer evidence to exact document locations (page/slide/row), making answers verifiable AI/RAG · Backend · 2025.12 ~ 2026.01 -
Generative AI (K-Bot) Vector DB Structure Design & Embedding Migration
A partition-key multi-tenant design and an embedding-model swap that never stops the service
Replaced per-customer/per-file collection growth with model/dimension-based shared collections, capping physical resource growth Backend · Database · 2025.11 ~ 2026.04
I treated a model replacement as a data migration, not a settings change.
The old service stays live while new embeddings are prepared, and traffic moves only after verification.
Read the full case study- Prepare Create collection
- Migrate Batch re-embed
- Verify Check count + consistency
- Apply Switch service
- Observe Track failure + recovery
Not just the tools, but where and how I used them.
Each capability links to real project evidence so you can inspect the implementation scope and reasoning.
I design the preprocessing, chunking, embedding, and citation flows that drive RAG retrieval quality inside real products.
From data analysis to production AI systems.
I started by narrowing problems with data. Today I build the backend and RAG systems behind production generative AI.
View full experience-
AI Engineer · Generative AI (K-Bot)
Developed the backend of Generative AI (K-Bot) in Younglimwon Soft Lab's K-System AI, building the full RAG pipeline from document preprocessing to Vector DB structure and metadata design and RAG answer source tracking. I also built MCP servers that let AI agents perform Google Workspace and Microsoft 365 tasks from natural language alone. -
Data Analysis Intern · SCR Process Data Diagnosis
Owned data processing and analysis (30% contribution) on an LS Cable project diagnosing data readiness for SCR wire-breakage factors and process conditions. Analyzed process data to identify crack root causes and patterns, and proposed manufacturing improvements with a baseline prediction model.
Notes left while solving problems
- ↗
What I Check First When Debugging RAG Migrations
Migration bugs become clearer when status, deleted documents, and apply paths are separated.
- ↗
A Portfolio Needs Proof Before Decoration
A developer portfolio should show traces of problem-solving before pretty copy.
- ↗
The Order I Use to Narrow Deployment Bugs
Logs, health checks, auth, and direct repros narrow causes faster when checked in a fixed order.
I turn complex systems into structures teams can understand and operate.
I go beyond shipping a feature. I document reproduction, boundary design, deployment, and verification as decisions that can be reused.
I started out in data analysis and now build Generative AI (K-Bot), a production AI feature in K-System AI, at Younglimwon Soft Lab. I work mainly on Python/FastAPI backends, RAG retrieval with Vector DBs and embeddings, and Azure-based data pipelines — and when needed I build the frontend myself with Svelte and TypeScript.
More about me and my workAsk a question more specific than a résumé can answer.
Answers use only the projects, experience, and technical notes published here.