Academic Homepage

Junwon Seo

Undergraduate student at Seoul National University studying Civil & Environmental Engineering and Artificial Intelligence.

I am interested in LLM agents for automated research.

Research Interest

Long-term research interests and current focus.

I am interested in LLM agents for automated research, particularly how they maintain and update state and make sequential decisions over long-horizon tasks. My current research focuses on agentic retrieval and dynamic retriever routing.

Education

Current undergraduate program and relevant coursework.

Seoul National University

Mar 2024 - Present

B.S. in Civil & Environmental Engineering; Interdisciplinary Major in Artificial Intelligence

  • Relevant coursework: Machine Learning Fundamentals, Deep Learning Fundamentals, Data Structures & Algorithms, ML Math & Programming Practice

Research Experience

Academic lab experience and research-facing work.

LDI Lab, Seoul National University

Jun 2026 - Present

Undergraduate Research Intern · Research Mentor: Dohyeon Lee

  • Investigating adaptive retriever selection for reasoning-intensive retrieval.

Smart Construction Lab, Seoul National University

Dec 2025 - Feb 2026

Undergraduate Research Intern · Advisor: Prof. Youngjib Ham

  • Designed state representations and action spaces for RL-based control of excavators and vibratory rollers, informed by literature analysis and practical constraints.

Selected Projects

Research tooling, agent infrastructure, and deployed AI products.

Research Loop

GitHub

Auditable autonomous experimentation plugin

  • Developed a Codex/Claude Code plugin and deterministic Python runner that extends open-source agent-skill scaffolding into approved execution and evaluation contracts.
  • Enforced Git-worktree isolation, plan-hash approval, metric validation, experiment budgets, and append-only research ledgers.

Margin

GitHub

Local-first research paper companion

  • Built an Electron desktop app for PDF ingestion, figure/table review, evidence-linked Markdown notes, and local SQLite persistence.
  • Integrated Codex App Server for language-preserving note organization while keeping papers and confirmed notes local.

Agent Skill Hub

GitHub

Local skill inventory and diagnostics app

  • Built a local web app that discovers and deduplicates skills across 59 AI coding-agent platform definitions and project roots.
  • Implemented snapshot diffs, H001-H015 health checks, conservative path boundaries, safe prompt generation, and a 157-test verification suite.

YouTube Skill

GitHub

Open-source video publication skill

  • Built Codex/Claude adapters and a shared Python engine for OAuth, resumable uploads, manifest validation, and publication safeguards.

InteriAI

Live site

Planning & Development · AI-powered interior design product

  • Built and deployed an AI web product that transforms room photos into multiple interior design styles through a prompt-driven image-generation workflow.
  • Implemented Google OAuth, credit-based SaaS usage, Korean/English multilingual support, and production deployment on Vercel.

PetStyle Studio

Live site

AI pet-fashion product

  • Supports more than 10,000 prompt-driven style combinations.

Awards & Activities

Selected academic and AI-building activities.

1st Place, KRAFTON Track — Cofathon: AI Native Battlegrounds

Jul 2026

KRAFTON & CJ Olive Young

Scoped a field problem through interviews with a field manager and shipped AFTER MATCH, a post-match interview practice web app for PUBG Esports players that coaches only player-confirmed Korean transcripts, with schema-validated LLM feedback and manual-input fallback when speech recognition fails.

Academic Excellence Scholarship

2025 Spring, 2025 Fall

Seoul National University

Kakao AI Top100 Finalist

Designed AI harnesses for prompt engineering, output verification, and practical LLM application workflows.

Skills

Technical stack and AI workflow tooling.

Languages & Frameworks

  • Python
  • JavaScript / TypeScript
  • Next.js, React, Tailwind CSS

AI / ML Systems

  • LLM agents
  • Information retrieval
  • Tool-use systems
  • LLM evaluation