Edge Data Collection
Collect real-world data using Jetson and Raspberry Pi devices in production environments.
AI Systems Engineer
Woohyuk Kang
Computer Vision & Edge AI Systems Engineer
I design and operate edge-to-cloud AI systems that collect, process, and operationalize real-world data.
What I Do
Computer Vision & ML
Build models optimized for real-world robustness, not just benchmark accuracy.
Backend API Systems
Design FastAPI systems for image ingestion, device authentication, and data pipelines.
Infrastructure & Deployment
Operate systems using Docker, Terraform, Cloudflare Tunnel, and VPN.
Applied Research
Analyze model failure (shortcut learning) and design mitigation strategies.
Architecture
Edge Devices
Jetson Orin / Raspberry Pi
FastAPI Server
Image upload / device authentication / encryption
Storage Layer
PostgreSQL / Redis / MinIO
Application Layer
Dashboard / Analytics / Model Retraining
3 Devices Deployed
30,000 Images Collected
Real-world Campus Deployment
Featured Projects
HTPeo Edge-to-Cloud AI Pipeline
Built a full AIoT system from edge devices to cloud infrastructure.
Tech: Jetson, FastAPI, PostgreSQL, MinIO, Docker, Terraform
Results:
- 3 devices deployed in campus environment
- 30,000 images collected in 2 weeks
- Pipeline evolved from HTTP → streaming → batch → edge inference
Shortcut Learning Research
Diagnosed why models fail on unseen objects.
Tech: PyTorch, ConvNeXt, K-means
Results:
- Accuracy improved from 69.43% to 73.44%
- Reduced bias in unseen combinations
SMCNet
Dual-label learning for better generalization.
Tech: YOLOv5, PyTorch
Results:
- +62.78% improvement in unseen object recognition
- Patent officially registered
Research
Problem:
Models fail on unseen objects.
Insight:
Models rely on shape/color instead of material (shortcut learning).
Approach:
- Latent space clustering (K-means)
- Bias detection using conditional probabilities
- Filtering, reweighting, feature masking
Result:
Improved accuracy and reduced variance in unseen conditions.
Awards
Minister of Environment Award
Ministry of Environment, South Korea · June 2025
Grand Prize in Environmental Data Utilization Competition
KCC Chairman Award
Korea Communications Commission · December 2025
Grand Prize in LBS Startup Challenge
Seoul Mayor Award
Seoul Metropolitan Government · October 2024
Top Prize in Smart Life AIoT Hackathon
Patent Registered
Korean Intellectual Property Office · 2022
Dual-label waste classification model (SMCNet) officially registered
SW Mileage Ranking #1
Kyung Hee University · 2024
Ranked 1st among all software students
About
I am an engineer focused on real-world systems, not just models.
I care about how systems behave after deployment, not just during training.
I design systems end-to-end, from data collection to infrastructure.
I iterate based on failure and continuously improve systems.
My goal is to build AI systems that actually work in real environments.
Contact
Email: woohyuk@khu.ac.kr
GitHub: github.com/lavaskiller
Open to Edge AI / ML Systems / AI Backend roles.