A researcher who connects the theoretical principles of mathematics with the practical applications of AI models.

A researcher with a passion for using mathematics to enhance AI model’s reasoning capabilities and reliability, while also exploring novel and efficient architectures.

I often enjoy solving diverse algorithmic problems through coding challenges and find satisfaction in simplifying complex issues.

Research Experience

• 2023.07 - 2023.12 | AI Grand Challenge 2023
  Utilized the openai-API with GPT-3.5 to generate and refine a dataset for fine-tuning a model on multi-answer questions, leading our team to a 7th place finish.

• 2024.02 - 03 | Practicing RAG
  Researched Retrieval-Augmented Generation (RAG) and implemented a model for the DACON competition, “HansolDeco Wallpaper Q&A.”

• 2024.03 - 06 | KCC 2024 and Hallucination
  Classified hallucination types in QA tasks into five categories. My research, published at KCC 2024, identified that natural language evaluation metrics like BLEU, METEOR, and ROUGE have distinct strengths and weaknesses in detecting specific types of hallucinations.

• 2024.07 - 2025.01 | Internship at SK Magic
  As a data scientist intern, I focused on product planning and development, researching on-device LLM ecosystems and data infrastructure for new products while developing corresponding algorithms.

• 2025.03 - PRESENT | AI and Math
  Currently designing methods for AI to understand and verify mathematical problems through formalized languages and algebraic structures.

Research Focus

My research centers on leveraging mathematical theory to enhance the reliability and reduce hallucination in AI models. I’m particularly focused on addressing the limitations of existing tokenizers’ embedding spaces through an algebraic approach. Specifically, I am designing a novel numerical token embedding space that preserves algebraic meaning by utilizing algebraic structures such as the Adele ring. My goal is to fundamentally improve model architectures with mathematical rigor, enabling AI to reason more deeply and logically about data.

Research Goal

My ultimate goal is to pioneer AI models capable of rigorous mathematical reasoning, which can fundamentally advance the field of artificial intelligence itself. While current LLMs demonstrate remarkable linguistic abilities, they often lack the foundational logic required for mathematical tasks. By implementing a new embedding space built on algebraic principles, I aim to create AI systems that move beyond mere pattern recognition and engage in logical, systematic thinking. This research is not just about solving math problems; it’s about building a new class of AI that can perform verifiable and reliable reasoning. This will pave the way for highly trustworthy AI applications in critical domains, setting a new standard for how we design and evaluate intelligent systems.