$ whoami

Jeonghoon Park

M.S.-Ph.D. StudentUNIST

$ about

I am an M.S.-Ph.D. student in Computer Science and Engineering at Ulsan National Institute of Science and Technology (UNIST). At the Ubiquitous AI Lab, I am advised by Prof. Taesik Gong.

$ research interests

As AI models become more capable, their computational demands also grow. My research in Efficient AI focuses on reducing memory usage and inference latency. Building on this, I study On-device AI to address the challenges of running these models within the resource constraints of users' own devices. I also work on Personalized AI, adapting general-purpose models to individual needs and preferences.

$ cat profile.png

$ tail -n 3 news.log

2026.09

Started UNIST’s integrated M.S.-Ph.D. program under Prof. Taesik Gong at UAI Lab.

2026.08

Graduated magna cum laude with a B.S. in CSE from UNIST, one semester early (7 semesters).

2026.07

Attended ICML 2026 (Jul 7–9) in Korea, my first opportunity to experience an academic conference.

Experience

Ubiquitous AI Lab – Ulsan, Republic of Korea

Graduate Student Researcher

2026.09 – Present

Advisor: Prof. Taesik Gong
Research on on-device and efficient AI systems

Undergraduate Research Intern

2025.01 – 2026.08

Advisor: Prof. Taesik Gong
Research on efficient VLMs and agentic systems

Topmath – Incheon, Republic of Korea

Mathematics Instructor

2021.07 – 2024.07

Taught and managed math classes for middle and high school students.

Education

UNIST – Ulsan, Republic of Korea

Integrated M.S.-Ph.D. in Computer Science & EngineeringIntegrated M.S.-Ph.D. in CSE

2026.09 – Present

Government Scholarship (Full Tuition Waiver)

B.S. in Computer Science & EngineeringB.S. in CSE

2020.03 – 2026.08

Full-tuition scholarship for all semesters (~US$20K total)
Graduated one semester early

Publications

*:Co-primary authors; †:Co-corresponding authors

SharpDraft: Accelerating Long-Context Speculative Decoding with Cardinality-Aware Query Scaling
2026.10
Jeonghoon Park*, Seongwoon Jo*, Jongwon Lee*, Taesik Gong
Under review at a top-tier conference
PDF arXiv
Global-Local Synergy for Efficient Visual Grounding in GUI Agents
2026.01
Jeonghoon Park, Mingyu Kim, Hojun Lee, Taesik Gong
Under review at a top-tier conference

Awards & Honors

Magna Cum Laude
2026.08
Graduated magna cum laude with a B.S. in Computer Science and Engineering from UNIST
Academic Excellence Award
2026.08
Recognized for a semester GPA of at least 4.0 out of 4.3
Academic Excellence Award
2025.08
Recognized for a semester GPA of at least 4.0 out of 4.3
Academic Excellence Award
2024.12
Recognized for a semester GPA of at least 4.0 out of 4.3

Patents

Method for grounding GUI based on Vision-Language Model
2026.09.21
Taesik Gong, Jeonghoon Park, Mingyu Kim, Hojun Lee
Korean Patent Application No. 10-2026-0180383

Teaching

NOVA 508: On-Device AI for Smart Manufacturing TA
2026.09 – 2026.12
Ulsan AI Novatus Academia (9th) PBL TA
2026.09 – 2026.11
Hyundai Heavy Industries AI Novatus Academia (1st) Theory Education TA
2026.08
Ulsan AI Novatus Academia (9th) Theory Education TA
2026.08
Busan AI Novatus Academia (1st) Theory Education TA
2026.07
Gyeongnam AI Novatus Academia (7th) Theory Education TA
2026.06
LG Electronics Living DX Course Theory Education
2026.01
Gyeongnam AI Novatus Academia (6th) PBL TA
2025.06 – 2025.09
Ulsan AI Novatus Academia (8th) Theory Education TA
2025.07
LG Electronics Living DX Course Theory Education TA
2025.07
Gyeongnam AI Novatus Academia (6th) Theory Education TA
2025.05
LG Electronics Living DX Course PBL TA
2025.02 – 2025.03
LG Electronics Living DX Course Theory Education TA
2025.01

Projects

UbiClaw
2026.03 – 2026.06
Built SLO-aware benchmarks and polite LLM throttling on Apple Silicon.
ConstraintRAG
2026.05 – 2026.06
Built a constraint-aware retrieval-augmented QA pipeline for Natural Questions.
doYOLOngo
2026.04
Built an on-device, voice-driven YOLO detector for iOS.
HARmony
2026.04
Built a real-time Human Activity Recognition (HAR) app for iOS.
SKiP: SVM with KNN and Probabilistic Weighting
2025.10 – 2025.12
Built a noise-robust SVM using KNN and probabilistic weighting.
PIGG: Personalized Interactive GUI Grounding
2025.10 – 2025.12
Built a GUI agent VLM with LoRA-based fine-tuning for user-specific adaptation.
CosmicMiner Robotics
2025.09 – 2025.12
Developed a multi-drone system with autonomous coordination and path planning.
Pintos Project
2025.03 – 2025.06
Implemented operating system components in Stanford's Pintos.
Traveling Salesman Problem (TSP) Solver
2025.05 – 2025.06
Implemented TSP algorithms and proposed an MCMF-based heuristic.
Sorting Algorithm Analysis
2025.03 – 2025.04
Implemented, benchmarked, and analyzed multiple sorting algorithms.
TinyLLM
2025.01 – 2025.02
Explored efficient LLM architectures for on-device environments.
One-Year Problem Solving Challenge
2024.01 – 2024.12
Solved algorithmic problems every day for 366 days, completing 1,000+ problems in 2024.

Else

$ cat mbti.json

ESTJ

I'm a strong ESTJ. I enjoy meeting new people and approach things with practicality, structure, and optimism. Feel free to reach out; I'm easy to talk to.

E99%

ExtravertedIntroverted

S72%

SensingIntuitive

T89%

ThinkingFeeling

J95%

JudgingProspecting

$ cd band/

Performing with UNIST's band club Unplugged on keyboard, with occasional vocals. But now, I'm practicing guitar.

$ ls photos/