Chaeho Yu

Natural & Life Sciences

Software Department

An independent research project motivated by the observation that vector search speed and memory usage can bottleneck RAG system response performance. I designed a neural network-based dimensionality reduction model that preserves both distance relationships and KNN neighbor rankings while reducing embedding dimensions.

Reduced 768-dimensional embeddings to 384 dimensions while maintaining ~89.9% search result consistency on FAISS HNSW indices, achieving ~1.5x faster search speed and ~48.2% memory reduction. Started this research during high school, formalized it into a paper after university admission, and presented at KCC 2026 as my first academic conference presentation.

3-person team / Team Lead / AI Development / Backend / UI·UX Design

An AI-powered paper platform born from the difficulty of quickly finding and understanding important papers among the flood of daily publications. Designed a system that collects arXiv papers, filters them by criteria, and transforms them into news-format content for users.

Responsible for the paper collection/filtering/news pipeline, recommendation features, backend architecture, and full UI/UX design. Inspired by The AI Scientist, introduced a structure where LLMs evaluate and select papers using self-reflection and ensemble approaches.

Team Project / Core Idea & Planning / AI·CV Development

A project addressing how existing ticketing platforms are easily manipulated by automation tools like Selenium. Focused on creating a ticketing structure where fraudulent purchases are inherently difficult, rather than blocking scalping after the fact.

Proposed the Selenium-Guard concept making HTML elements hard for automation to target, developed an analog clock CAPTCHA readable by humans but difficult for CV models, and researched iPhone TrueDepth camera-based facial recognition for mobile ticket eKYC.

2-person team / Team Lead / 70% contribution

A project observing behavioral patterns when LLM-based agents can directly manipulate computer systems. While existing LLM safety discussions focus on response control, we identified a gap in understanding actual system-level "actions" and designed experiments accordingly.

Built Linux environments on Raspberry Pi and VMs, enabling LLMs to execute commands via SSH. Observed behavior patterns including system info gathering, updates, security hardening, and web server deployment. Responsible for project planning, experiment design, instance setup, SSH I/O control module, and self-reflection prompt architecture.

Personal/Team Project / Planning / Design / Backend Development

An AI chat interface aimed at making users feel AI is a friend rather than a simple chatbot. Generated conversations with LLMs, synthesized voice with VITS, and applied Live2D character animations. Developed as a React + Electron desktop application, handling overall planning, full design, and backend development.

Scholub — AI Paper Platform

Developed an AI platform that collects, filters arXiv papers and transforms them into news content. Led project planning, AI development, backend, UI/UX design as team lead.

VEDGE — Anti-Scalping Ticketing Platform

Developed a platform to prevent ticketing automation and scalping. Contributed Selenium-Guard concept, analog clock CAPTCHA, and TrueDepth-based facial recognition/eKYC research.

BunnySnooper2 - BlackDawn

Handled game design, UI, news system, building and character feature development in a defense game project.

DuDu — AI Assistant App for Borderline Intelligence

Developed an AI character-based app to assist daily life for people with borderline intelligence. Handled LLM features and app backend development.

Good Night Good Dream

Handled items, HTTP communication features, presentation materials and presentation in a hyper-casual game project.

MonsCure

Developed a 2D RPG reinterpreting the theme of "health" as monster health in-game. Handled first stage, player movement, Git collaboration management, and presentation materials.

Founded & Operated / Teaching / Project Mentoring

Founded and operated the AI R&D club PARA, recognizing the lack of a deep AI learning environment at school. Managed all aspects including demand surveys, promotion, branding, website creation, interviews, member selection, and curriculum design.

Conducted all club lectures. Mentored junior students with limited development experience to complete personal projects. The club was later promoted to an official department-affiliated club.

Expanding into cybersecurity after university admission, studying C language, computer architecture, web, and network security fundamentals. Extending existing AI and service development experience toward system architecture and security perspectives.

Education Event Planning & Operations

Led software education events for middle school students including textbook writing, presentation creation, teaching, and hands-on project development. Taught Unity game development in 2023 and LLM chat platform development in 2024.

AI / LLM / Backend Education

Taught PARA members topics including Python, AI fundamentals, NLP, LLM usage, prompt engineering, fine-tuning, FastAPI, and MongoDB. Designed curriculum and materials from scratch so even beginners could produce working results.