I am a doctoral researcher in Artificial Intelligence at Yonsei University, working on LLM reasoning and post-training. My research focuses on how language models can acquire and improve reasoning capabilities through better training signals, feedback, and iterative refinement.

My work spans reasoning model training, reinforcement-learning-based post-training, knowledge distillation, synthetic and model-generated data, self-reflection, latent reasoning, and feedback-driven self-correction. My earlier research also includes structured and formal language generation, such as Text-to-SQL and molecular sequence generation.

I expect to complete my Ph.D. in February 2027 and am interested in Research Engineer, Research Scientist, and Applied Scientist opportunities involving model training, reasoning, post-training, and model capability improvement.

Research

LLM Reasoning & Post-Training

• GeneralThinker — RL-based post-training for domain-general reasoning with dense reward signals and fine-grained credit assignment.
• SpiralThinker — Iterative latent reasoning through text–latent interleaving and progressive representation alignment.
• TinyThiner — Reasoning knowledge distillation with synthetic reasoning data and iterative self-reflection.

Structured / Formal Language Generation

• LitE-SQL — Lightweight Text-to-SQL with supervised fine-tuning, execution-guided preference optimization, and self-correction.
• SELF-EdiT — Transformer-based molecular editing with SELFIES for structure-constrained molecular optimization.

Technical Focus

Model Training & Post-Training: SFT · DPO · GRPO-style RL · Knowledge Distillation · Synthetic Data · Self-Reflection · Reward Design · Model Evaluation

Reasoning: Language Reasoning · Latent Reasoning · Feedback-Driven Self-Correction · Test-Time Reasoning

Tools: PyTorch · Hugging Face Transformers · TRL · vLLM · Weights & Biases · Multi-GPU Training