↗fig. 01From clinical video to model validation↗fig. 02Comparing representations: ROC-AUC
Research · Vision AI · Veterinary Medicine
Canine Seizure Detection
I research vision AI that identifies potential seizure episodes in clinical videos, helping reduce gaps in seizure records for dogs with epilepsy.
↗42 dogs · 350 clips
·ROC-AUC 0.865
·First-author manuscript
note: review candidates, not automatic alerts →
Open this field note ＋
Jan 2026 – present · AI Rookie: 100 teams → 40 teams → Ongoing
I research vision AI that identifies potential seizure episodes in clinical videos, helping reduce gaps in seizure records for dogs with epilepsy.
350 clinical video clips from 42 dogs at Signature Animal Medical Center
Personally segmented and labeled videos under veterinary supervision
Used patient-level data splits to evaluate generalization to unseen dogs
Structured veterinary assessment criteria into 24 clinical questions to extract VLM features
Clinical knowledge-based representations achieved ROC-AUC 0.865, versus 0.771 for pretrained video representations
Evaluated a location-based baseline and collection bias from hospital backgrounds and staff intervention
With 79% sensitivity and 0.39 precision, prioritized flagging segments for review in long videos rather than automated alerts
Role · Problem definition, dataset creation and labeling, model design and experiments, error analysis, and first-author manuscript preparation
FIELD NOTE 02
↗fig. 01VetSync, in the hospital workflow↗fig. 02A closer look at inpatient management
Work · Product · Healthcare
VetSync
An inpatient management system used by 459 veterinary professionals. I lead frontend development for VetSync, used at major animal hospitals in Korea including Western Animal Hospital and VIP Animal Hospital. I joined VetU1 as its third early member.
↗459 clinical users
·Frontend lead
·Real hospital workflow
note: used by 459 clinicians ↗
Open this field note ＋
Jan 2026 – present | Frontend Lead · Vision AI / Software Developer · Transitioned to full-time in Jul 2026
An inpatient management system used by 459 veterinary professionals. I lead frontend development for VetSync, used at major animal hospitals in Korea including Western Animal Hospital and VIP Animal Hospital. I joined VetU1 as its third early member.
Develop and improve inpatient management features used by clinical staff
Explore bringing vision AI research into hospital workflows
Translate clinical workflows, data structures, and feature requirements into working software
I care about understanding why something works—and checking whether it really does.
01
VISION AI / CLINICAL REPRESENTATION
Canine Seizure Detection
First-author manuscript in preparation
VLM-based clinical representation
ROC-AUC 0.865
Research notes ＋
First Author · Manuscript in preparation
Seizure detection research using 350 clinical videos from 42 dogs and VLM-based clinical features. Evaluation covers model performance, data bias, baselines, and practical clinical use cases.
Surveyed veterinarians and technicians at five referral hospitals using the same inpatient system to assess suitability for digitization, changes in efficiency, and usability issues.
118 respondents: 67 veterinarians and 51 technicians
Treatment orders, completion tracking, and daily observation records were rated well suited to digitization
Key limitations included handoffs, patient prioritization, rigid input fields, and system stability
↗fig. 01Inpatient care digitization · conference poster
conference03
MODEL VALIDATION / CLINICAL ML
CCDS / iConnectome
Model validation
Leakage analysis
Age deconfounding
Examining data leakage, inflated performance, and the effects of age in existing CCDS models.
ongoing
One more note: speech-based Alzheimer’s disease research ＋
Age obscures acoustic signal in Alzheimer's disease speech classification
Siu Beom & Sun-shin Yi · Konkuk University
Research on how age as a confounding variable affects speech-based Alzheimer’s disease classification. Submission to npj Digital Medicine is planned.
PAGES ALONG THE WAY
How I got here.
04↗
2024
a new chapter
Started veterinary medicine at Konkuk
2025
learning by making
Android Developer @ NyamCoach
Lead Developer @ RISE
Hult Prize Campus Director
2026
connecting the dots
Joined a veterinary research lab
Jan · Joined VetU1 as its 3rd early member
Jul · Transitioned to full-time at VetU1
Konkuk University English Newspaper · Head of External Relations
AI Rookie finalist
Conquer Health · 2nd
Try Everything · Grand Prize
First-author research
hwpx-builder
All experience & activities ＋
VetU1 · Vision AI / Software Developer · Joined Jan 2026 · Transitioned to full-time Jul 2026
Konkuk University · Undergraduate Researcher · Jan 2026 – present
Konkuk University RISE Project Team · Lead Developer · Jun – Nov 2025
Hult Prize · Campus Director · Aug 2025 – Feb 2026
Konkuk University English Newspaper · Head of External Relations
