I’m a software engineer based in South Korea with 12 years of experience under my belt "I’ve spent my career obsessing over reliability in every project, whether it’s B2B SaaS, high-traffic real-time services, or robust data pipelines.

When I’m not deep in the code, you’ll likely find me running or meditating. These practices help me keep a clear perspective and stay grounded when engineering challenges get complex.

Recently, I have been focusing on LLM-based product development and studying the foundational principles of LLMs. Integrating AI into my development workflow has significantly improved my productivity, and I am now expanding my focus into MLOps and Research Engineering to build efficient, AI-native infrastructure.

An open-source core engine for Tail Villain, an AI-powered interview and career roadmap platform. I am currently focusing on building highly efficient, production-ready AI infrastructure.

• Embedded RAG via pgvector: Built a retrieval-augmented generation (RAG) pipeline directly within a PostgreSQL environment using pgvector. This enables real-time extraction of personalized follow-up questions by vectorizing user interview history and resumes.
• Real-time AI Audio Streaming: Engineered a WebSocket-based streaming layer for Gemini Multimodal Live API. Resolved latency and STT desynchronization issues to ensure a seamless, low-latency conversational experience.
• Prompt Optimization & Context Management: Implemented Prompt Caching and sliding window-based context management to drastically reduce token costs and prevent context degradation in long-turn multi-turn conversations.

I keep my problem-solving skills in shape by solving LeetCode problems. It’s a great way to stay consistent and refine my approach to complex logic.

• 👉 View my study repo and PR history

A space where I document my journey into MLOps, deep technical troubleshooting, and architectural decision-making.

• Blog: codetosoul.com