한국어 보기
👋 Introduce
Introduce
- Specialized in building AI agents using LangChain/LangGraph, with hands-on expertise in Vector RAG and Graph RAG.
- Experienced in designing and implementing RAG pipelines for unstructured data using AI models in on-premises environments.
- Improved product quality through performance optimization and device compatibility improvements for B2B products including Ontovia, NexPlayerSDK, and AgensSQL.
- Provided technical support to global customers, solving complex technical issues and developing strong problem-solving expertise.
- Focused on systematically analyzing and utilizing data to produce accurate and high-quality results.
- Proactively adapt to new challenges and changes, solving diverse software development issues through creative and multi-dimensional approaches.
Contact & Channel
⚒️ Skill
Engineering
C, C++, Android, iOS(Objective-C, Swift), Postgres Internal, Shell script
DevOps
Jenkins
Tools
VS Code, Git, Bitbucket, Sourcetree, Jira, Slack
💼 Work Experience
Company name changed from Bitnine
2024.02 - Now
AI Solution Development Team
Core R&D Team
2023.05 - 2024.02
HWP Parser Development
- Implemented an HWP document extraction module based on rhwp.
- Designed and implemented structured document parsing logic optimized for LLM applications.
KIMFT(Korea Institute of Maritime and Fisheries Technology) — AI Search Assistant
AI Agent Engine Architecture & Development
- Architected a dedicated Vector RAG / Graph RAG agent architecture based on Ontovia.
- Designed and implemented agent orchestration for query preprocessing, RAG routing, Text-to-SQL, query reformulation, document parsing, and information extraction.
On-Premises GPU Server Infrastructure
- Configured an on-premises GPU server environment using vLLM and Hugging Face.
- LLM: gpt-oss-20b
- Embedding Model: Qwen3-Embedding-8B
- Reranker Model: Qwen3-Reranker-8B
- Optimized memory usage for individual AI models.
Unstructured Document Analysis & Parsing Strategy
- Designed document processing strategies for PDF, HWP, HWPX, DOC, DOCX, PPT, PPTX, XLS, and XLSX.
- Implemented pgvector-based vector embedding and retrieval.
Ontovia
AI Agent Engine Architecture & Development
- Designed RDB RAG, Vector RAG, and Graph RAG AI agents using LangGraph.
- RDB RAG Generation & Retrieval: Designed and implemented Text-to-SQL functionality for PostgreSQL and Oracle.
- Vector RAG Generation & Retrieval: Designed and implemented pgvector-based embedding and retrieval pipelines.
- Graph RAG Generation & Retrieval: Designed and implemented AgensGraph-based Text-to-Cypher functionality.
- Natural Language Query Processing: Designed and implemented RAG routing, Query Reformulation, Few-shot prompting, and Guardrails.
GPU Server Infrastructure
- Built AI infrastructure across on-premises and cloud environments.
- Cloud GPU: Elice H100
- Local GPU Server: 4 × NVIDIA A6000
- Optimized AI model memory usage through model parameter tuning.
AI Model Deployment
- OpenAI: GPT-4o series, GPT-5.x series, text-embedding-3-small, text-embedding-3-large
- Google: Gemini 3.5 Flash
- Ollama: gpt-oss-120b, Qwen3-Embedding-8B, Qwen3-Reranker-8B
- vLLM: gpt-oss-20b, Gemma 4 27B, Qwen3-Embedding-8B, Qwen3-Reranker-8B
Disclosure IR MCP
- Orchestrated MCP servers for corporate disclosure data.
- Improved data search capabilities.
AgensSQL — PostgreSQL-based DBMS
Customer Technical Support
- Provided product installation and licensing support for customers.
Oracle Query Compatibility
Improved database migration compatibility by supporting Oracle-specific SQL syntax.
REVOKE CONNECT FROM RoleSpec
GRANT CONNECT TO RoleSpec
ALTER SESSION SET <set>
Oracle Function Compatibility
Improved database migration compatibility by matching the behavior of Oracle-specific functions.
- Added support for Oracle-specific date formats such as
FF7 and FF9.
- Added support for Redwood (EDB) date styles.
- Added support for Oracle-specific data type functions.
AgensSQL — Product Licensing Module
Product License File Integration
- Implemented product license validation and application logic.
- Prevented unauthorized use of the product by customers without valid licenses.
- Improved product commercial readiness.
agenscrypto Extension
Korean Cryptographic Algorithms
- Implemented ARIA and SEED, Korean-developed cryptographic algorithms.
- Developed a column-level data encryption extension.
- Added support for additional domestic security algorithms.
AHM Watchdog
High Availability Submodule for Agens Products
- Implemented a server monitoring submodule for uninterrupted HA operation.
- Implemented server recovery functionality to improve system stability.
💻 Personal Experience