지승현Seunghyun Ji

standingbehindnv@gmail.com·GitHub·YouTube

Knowledge gains its value when shared. I find what should be known, share with diverse perspectives, and implement — as an AI Research Architect.

Highlights

Continuous follow-up on cutting-edge research and meta-analysisPaper notes →

Intuitive explanations to bridge perspectives across disciplinesYouTube lectures →

Solo development of the AI OS project (yaar) and other AI workGitHub Repo →

Experience

IT Consultant @ VAIIM Consulting Group

25.09 – 26.03

On-premise LLM R&D project for vertical domain at automotive company H
On-premise LLM build project for a KEPCO-affiliated organization

Head of AI Research & Development @ Rivetta Inc.

25.03 – 25.06

Multi-Agent System R&D
Platform development, QA, and optimization using Codex

AI Team (reporting to CTO) @ Connectwave Inc.

24.07 – 24.10

Worked on the in-house LLM project (PLAi) [Project doc]
R&D on item representation [Project doc]

NLP Researcher, AI Research Team (DS Division) @ Telepix Inc.

23.03 – 23.10

Implemented and improved various models: filtering, translation, sentence embeddings [Public repo]
Authored and presented a paper on translation data collection methodology [Paper] [Slides]
Implemented vision models for cloud detection; led seminars on recent research [Seminar slides]

Education

Soongsil University Graduate School M.S. in Software Engineering, GPA 4.36/4.5

21.03 – 23.02

Thesis: Masked Language Modeling using Text Length (advisor: Prof. Soowon Lee) [Paper]
Industry-academia R&D project [Slides]

Soongsil University B.S. in Computer Science, GPA 3.72/4.5

14.03 – 21.02

Built and published the AI sentiment-diary app 'Weather of the Heart'
Coursework and short papers in Korean Linguistics [On cognitive models at the concrete level] [The principle of focus]

Publications

Seunghyun Ji. Ortho-Hydra: Orthogonalized Experts for DiT LoRA. Arxiv preprint. [Link]

Seunghyun Ji, Sinulangga H.R., and Darongsae Kwon. Robust Guidance for Unsupervised Data Selection: Capturing Perplexing Named Entities for Domain Specific Machine Translation. SIGUL 2024 (LREC-COLING 2024 satellite workshop). [Link]

Seunghyun Ji. Confidence Regularized Masked Language Modeling using Text Length. M.S. Thesis, Soongsil University. [Link]

Seunghyun Ji, Soowon Lee. Substitute Word Extraction via a Replaced-Token Detection Model. KIISE Journal. [Link]

Seunghyun Ji, Soowon Lee. Substitute Word Extraction via a Discriminative Model. KCC 2022. [Link]

Seunghyun Ji, Youngmo Kim, Seok-Yoon Kim. A study on performance enhancement of bi-LSTM text sentiment model using bootstrap method with limited tokens. EEECS 2020.

Skills & Activities

Tech Stack

End-to-end deep learning model dev/training experience including PyTorch and CUDA customization
Python, C, C++, Java
Cloud computing on GCP/AWS VMs and remote training experience
Git, Bun/uv, postgre and other deployment/DBMS tooling

Conference Reviewing

SIGUL 2024 (LREC-COLING 2024) workshop — Program Committee [Name list]
EMNLP 2023 — Invited Reviewer
ACL 2023 — Emergency Reviewer (≤3 papers)

Consulting & Advisory Activities

KOCCA invited lecture: 'Development using Claude'
Invited AI expert for KRIT DTiMS 4.0 advancement project audit
Project consulting and mentoring [Featured project]
AX consulting @ IBS Co., Ltd. (based in Japan)

Teaching & Study Activities

Inflearn free course [Course page]
YouTube lectures [Channel]
Initiative: 'Make recent research easy and fun to read' [Project page]

Competitions

2022 NIKL AI Capability Evaluation: Aspect-Based Sentiment Analysis [GitHub]

Other Experience

NLP AI Researcher (short-term contract) @ IBS Inc., Japan23.11 – 23.12

TOEIC 8852022.01

Curriculum design assistant, Soongsil Digital Innovation Talent Institute2022/01 – 2022/02

Soongsil debate club 'Manjang-ilchi' member2020.09 – 2021.12

Call center contractor in Japan (Linkage Service Co., Ltd.)2018/09 – 2019/05

JLPT N12016.08

Conscripted police service, completed2015.11 – 2017.08