Minha Kim

AI Engineer @ Nota AI · Seoul, South Korea

About

I am an AI Engineer at Nota AI, working on vision-language models (VLM) and computer-vision systems for intelligent transportation — large-scale model evaluation, automated prompt optimization, and edge-deployable detection pipelines.

Previously, my research focused on deepfake detection and face anti-spoofing with continual learning and knowledge distillation (ACMMM, NeurIPS Datasets & Benchmarks, CVPRW). I enjoy taking models from research benchmarks to reliable field deployment.

News

Mar 2026Our paper on domain-generalizable face anti-spoofing is accepted at Pattern Recognition.
May 2024Joined Nota AI as an AI Engineer.
Dec 20225th place at the SNUH Medical AI Challenge.
Nov 2022RCRL accepted at AVSS 2022 as an oral paper.
Jan 2022FakeAVCeleb accepted at NeurIPS 2021 Datasets & Benchmarks Track.
Oct 2021CoReD accepted at ACMMM 2021 as an oral paper.
Jul 20211st & 3rd place at the IITP AI Grand Challenge (fallen-person detection in CCTV).

Experience

Nota AI — AI Engineer May 2024 – Present

Seoul, South Korea

Built VLM-based fire/incident detection for intelligent transportation systems: led large-scale benchmarking across model families, quantization schemes, and input resolutions, and designed a CLIP-filtering + VLM ensemble pipeline that substantially reduced false alarms in field deployment.
Developed a LangGraph-based automated prompt-optimization pipeline (failure-case mining → pattern analysis → automatic prompt rewriting → regression validation).
Solo-designed a classical-CV anomaly-trigger pipeline for PTZ cameras (background subtraction with grid- and rule-based logic), through evaluation-metric design and handover.
Conducted detection-model failure analysis at 100K+ frame scale and a QLoRA fine-tuning feasibility study for false-positive suppression.
Built gender-classification and age-estimation models for demographic-aware ad display on tablet ordering devices: constructed diverse training datasets, improved in-field accuracy, and authored data-collection guidelines with expansion-scenario plans (2024–2025).
Provided technical advising on model improvement to an overseas engineering team.
Anti-spoofing detection for smart door locks: designed the data-collection pipeline, analyzed failure cases under adverse conditions, and improved generalization for NXP edge deployment (2024).

NAVER Cloud — Research Intern Oct 2022 – Apr 2023

Seongnam, South Korea

Contributed to a vision anomaly-detection API: dataset design, condition-aware filtering, and a switch from single-frame to multi-frame prediction for output stabilization.
Research on generalizable face anti-spoofing: patch-based multi-task learning of local textures and texture-conversion augmentation.

DDS — Researcher Sep 2017 – Feb 2020

Seoul, South Korea

Developed 3D CAD algorithms for dental prosthesis design (VTK/OpenGL): real-time bridge generation, interactive preview by prosthesis type, and a 3D-axis correction tool.

Publications

Domain-generalizable Face Anti-Spoofing with Patch-based Multi-tasking and Artifact Pattern Conversion

Pattern Recognition (Elsevier), 2026 · co-first author

[paper]

RCRL: Replay-based Continual Representation Learning in Multi-task Super-ResolutionOral

AVSS 2022 · co-author

[code]

FakeAVCeleb: A Novel Audio-Video Multimodal Deepfake Dataset

NeurIPS 2021 Datasets & Benchmarks Track · 3rd author

[paper] [code]

Efficient Multi-Scale Feature Generation Adaptive Network

CIKM 2021 · 2nd author

CoReD: Generalizing Fake Media Detection with Continual Representation using DistillationOral

ACMMM 2021 · first author

[paper] [code]

FReTAL: Generalizing Deepfake Detection using Knowledge Distillation and Representation Learning

CVPR Workshops 2021 · first author

[paper]

Tear Extraction from Ultrasonic Images of Shoulder using Fuzzy Stretching and SOM-based QuantizationBest Paper

KIICE 2017 (domestic) · 2nd author

Tear Extraction from Ultrasonic Images of Shoulder Tendon using Image Processing

KIICE 2016 (domestic) · first author

Awards

1st place (2021.01) & 3rd place (2021.07), IITP AI Grand Challenge — fallen-person detection in CCTV
2nd place, DACON Hand-Gesture Recognition AI Competition, 2021
Top 1%, DACON · LG Camera Image Quality Enhancement AI Competition, 2021
5th place, SNUH Medical AI Challenge, 2022

Education

Sungkyunkwan University Feb 2022

M.S. in Software · DASH Lab (Advisor: Prof. Simon S. Woo)

Silla University Aug 2017

B.S. in Computer Engineering · GPA 3.98/4.5