Jihwan Kim

I am an integrated M.S.–Ph.D. student in the DAVIAN Lab at KAIST AI, advised by Professor Jaegul Choo. My research interests lie in the evaluation of agents, speech and audio processing, and multilingual LLMs.

Every project I have worked on started as a modeling question and ended as a measurement question. So far that has meant building an evaluation suite for web browsing agents, building effective audio detection and contributing an open-source full-duplex audio pipeline, and synthesizing data that converts unverifiable multilingual RLVR tasks into verifiable ones.

Fun fact: I finally built something with my day6.ai.kr domain 🙂 — come and have a look!

Research Interests

Agent Evaluation
Speech & Audio Processing
Multilingual LLMs

News

Sep 2026 Two papers accepted to EMNLP 2026 — cross-lingual RLVR via code-switched task synthesis to the main conference, and a next-scene-prediction benchmark for omni-models to Findings.
May 2026 Two papers accepted to ACL 2026 — Adaptive Table Retrieval to Findings, and Sommelier to the Industry Track.
Mar 2026 New preprint on speech: Sommelier — scalable, open multi-turn audio pre-processing for full-duplex speech language models.
Jan 2026 LiveWeb-IE was accepted to ICLR 2026 as a poster.
Jan 2026 The HyperCLOVA X 8B Omni and 32B Think technical reports are out on arXiv.

Publications

2026

No image
From Unverifiable Domain Text to Cross-Lingual RLVR via Code-Switched Task Synthesis

Cheonbok Park*, Jihwan Kim*, Shinyoung Joo, Joosung Lee, Hwiyeol Jo, Jeonghoon Kim, Jaegul Choo (*Equal contribution)
EMNLP 2026
- #RLVR
- #Cross-lingual
- #Data Synthesis
[arXiv]
No image
Can Omni-models Imagine the Future? A Diagnostic Benchmark for Multimodal World Modeling

Junyeob Kim, Jihwan Kim, Minha Jhang, Sungwook Jeon, Hwiyeol Jo, Jeonghoon Kim, Sang-goo Lee, Joosung Lee
EMNLP 2026 Findings
- #Benchmark
- #Omni-models
- #World Modeling
[ACL Anthology]
No image
LiveWeb-IE: A Benchmark For Online Web Information Extraction

Seungbin Yang*, Jihwan Kim*, Jaemin Choi, Dongjin Kim, Soyoung Yang, ChaeHun Park, Jaegul Choo (*Equal contribution)
ICLR 2026
- #Benchmark
- #Web Agents
- #Information Extraction
[arXiv] [OpenReview] [code]
No image
Retrieve Only Relevant Tables Whether Few or Many: Adaptive Table Retrieval Method

Taehee Kim, Seungbin Yang, Jihwan Kim, Jaegul Choo
ACL 2026 Findings
- #Table Retrieval
- #Text-to-SQL
- #Adaptive Retrieval
[arXiv] [code]
No image
Sommelier: Scalable Open Multi-turn Audio Pre-processing for Full-duplex Speech Language Models

Kyudan Jung, Jihwan Kim, Soyoon Kim, Jeonghoon Kim, Jaegul Choo, Cheonbok Park
ACL 2026 Industry Track
- #Speech LM
- #Full-duplex
- #Data Pipeline
[project] [arXiv] [code]
No image
HyperCLOVA X 32B Think

NAVER Cloud HyperCLOVA X Team (including Jihwan Kim as a Residency Intern)
Tech Report 2026
- #Reasoning
- #Vision-Language
- #Korean
[arXiv] [model]
No image
HyperCLOVA X 8B Omni

NAVER Cloud HyperCLOVA X Team (including Jihwan Kim as a Residency Intern)
Tech Report 2026
- #Omnimodal
- #Speech
- #Any-to-any
[arXiv] [model]

Education

2024–Present

Integrated M.S.–Ph.D. in Artificial Intelligence

KAIST AI · DAVIAN Lab

Advisor: Prof. Jaegul Choo

Research focus: agent evaluation, speech & audio processing, multilingual LLMs
2018–2024

B.S. in Software Convergence

Sungkyunkwan University (SKKU)

Magna Cum Laude · GPA 4.11/4.5

Seah Haiam Scholarship (2022–2024) · Dean's List (Spring 2022)

Experience

2024.09–Present

Graduate Researcher

KAIST AI · DAVIAN Lab

Research on agent evaluation, speech & audio language models, and multilingual LLMs.
2025.10–2026.08

WBL Residency Intern

NAVER Cloud

Built a domain-specific cross-lingual RLVR data-synthesis method and an evaluation pipeline for Audio LLMs covering both understanding and generation; implemented multimodal benchmarking for HyperCLOVA X 8B Omni and contributed to the HyperCLOVA X technical reports.