Wonjoong Kim

I am a Ph.D. student in the Graduate School of Data Science at KAIST, where I am fortunate to be advised by Prof. Chanyoung Park.

I do research with my awesome colleagues at DSAIL (Data Science and Artificial Intelligence Lab).

---

🔬 Core Research Focus

Self-Evolving LLM Agents

I work on how LLM agents can be evaluated and improved from their own experience, so that an agent becomes more reliable on long, multi-step tasks without a person hand-tuning it after every failure.

Keywords: Agentic AI, Tool-Use Agents, Multi-Agentic System, Self-Evolving Agents, Harness Optimization

Key Focus:

• Trajectory Evaluation: Judging the whole reasoning trajectory of a tool-using agent, not only whether its final answer is correct.
• Multi-Turn Agent Optimization: Step-level reward signals that tell an agent which intermediate steps helped, instead of a single outcome reward at the end.
• Harness Optimization: Automatically improving the harness around the model (its prompts, tool configurations, and control logic) from the agent’s execution traces.