MLLM Safety Guardrails
Safety-oriented guardrail models for multimodal large language models, with emphasis on robust behavior under visual-language inputs.
Undergraduate Researcher · Computer Science
JungHun Kim
I study safer and more useful intelligent systems, with current interests in multimodal LLM safety guardrails, deep reinforcement learning, and generative handwriting systems.
Research Interests
Questions I am currently working around.
Deep Reinforcement Learning
Learning-based decision systems and experimental RL methods developed through undergraduate research practice.
Generative Handwriting Systems
Style-conditioned handwriting refinement using diffusion-based generative models connected to practical mobile-to-server systems.
Experience
Research and product-building timeline.
Undergraduate Research Intern, UTLLab
Conducting research on safety guardrail models for multimodal LLMs.
Undergraduate Research Intern, DMSLab
Worked on deep reinforcement learning and large language model research topics.
Co-Founder, Lett
Founded Team Lett and explored technical entrepreneurship across app and web services.
Selected Work
Projects at the boundary of models and usable software.
Inkly: AI-Based Handwriting Refinement Application
Inkly analyzes handwritten input and generates improved handwriting while preserving the writer's personal style. The project explores style-conditioned handwriting generation with diffusion-based generative models and a full-stack mobile-to-server architecture.
Pzzk: A Clean, Serverless iOS Calendar
Pzzk is a minimal iOS calendar designed around the concept of being the simplest calendar in the world. Its core implementation contribution was iCloud synchronization without operating a separate server.
RepositoryEducation
B.S. in Computer Science and Engineering
Konkuk University · 2021.03 - 2027.02 expected
Major GPA: 4.23 / 4.5 · TOEIC: 865
Technical Skills
- Python
- SQL
- PyTorch
- React
- Django
- Git
- Docker
Contact