SaaSLive
img-src.io
Developer-first image CDN
Transform images directly through the URL, automatically serve WebP and AVIF, and pay no egress fees. Built on Cloudflare infrastructure.
- Image CDN
- URL Transforms
- WebP / AVIF
- Zero Egress
- Cloudflare
INNOVATION THROUGH CODE
One engineer.AI at every layer.Software that ships.
Taehun Kim Software Lab is a one-person software company in South Korea. One founder sets the direction and works with AI agents from the first commit through production operations.
- 17+
- Software engineering since 2008
- 26+
- Cloud & AI credentials
- Golden
- Kubestronaut — all CNCF Kubernetes certifications
- 3
- Products in production
Founder
Taehun Kim
Founder & CEO
South Korea
Taehun Kim has spent 17+ years in software engineering. He began by writing a Linux network driver for a TCP/IP chip, then ported the Linux kernel to one of the first ARM64 single-board computers before moving into machine learning — sound recognition, ML infrastructure for autonomous driving, and esports analytics.
He is a Golden Kubestronaut, a distinction awarded for simultaneously holding the full set of CNCF Kubernetes certifications, and has earned 26+ cloud and AI credentials across AWS, Google Cloud, and CNCF. He has contributed code to the Linux kernel, TensorFlow, and Buildroot.
In January 2026, he founded Taehun Kim Software Lab around a simple premise: when AI is orchestrated well, one engineer can take a product from end to end — architecture, code, infrastructure, and customer support.
Operating model
One person, amplified by AI
A traditional product team divides work by role. Here, one person makes every product decision while AI agents help execute the design, code, and operations.
- 01
AI-Native Solo Company
AI takes on the work of a traditional team, but not the judgment behind it. One person owns product, engineering, and operations — and remains accountable for all three.
- 02
AI Orchestration
One person directs multiple AI agents across coding, code review, and debugging. Iterations that once took weeks can now be completed in days.
- 03
Vibe Coding
A product idea can move directly into production-ready code. Design, implementation, and testing happen through pair programming with AI, while the person accountable for the outcome approves every release.
- 04
End-to-End Ownership
One person handles architecture, infrastructure, releases, and support, eliminating handoff gaps. A shorter chain means faster feedback and faster fixes.
Expertise
Depth across the whole stack
Seventeen years of experience, from bare-metal boards to production ML platforms. That breadth makes a one-person software company possible.
Agentic AI & Orchestration
Agent systems that plan, use tools, and maintain state — including a LangGraph-based DevOps agent built for an esports ML platform.
AI & ML Engineering
Training, optimizing, and serving production models — including scene classification with 99% accuracy, OCR, and on-device inference using TFLite quantization.
Cloud & Kubernetes
AWS and Google Cloud architecture, plus on-premises and managed Kubernetes — backed by the full set of CNCF Kubernetes certifications.
DevOps, SRE & MLOps
GitOps delivery with Argo CD, training pipelines built with Kubeflow and MLflow, and FinOps work that reduced an esports platform’s cloud bill by 60%.
Embedded & Linux Systems
Bootloaders, kernel drivers, and board support packages — including a TCP/IP driver merged into the mainline Linux kernel and a kernel port for one of the first ARM64 single-board computers.
Full-Stack Product Development
Web and mobile apps, APIs, and edge infrastructure — everything needed to turn working code into a reliable service.
Projects
Built here. Run here.
From production SaaS to hackathon entries, every project is designed, shipped, and operated by one person working with AI.
SaaSLive
Hero3D
Prompt-generated 3D hero backgrounds
Describe a scene in plain language and get a production-ready WebGL background. It runs at 60 fps on mobile, adapts to light and dark themes, respects reduced-motion preferences, and integrates with any stack in one line of code.
- WebGL
- Prompt to 3D
- React
- Landing Pages
- 60 fps Mobile
Mobile AppLive
The Detective’s Notebook
A daily deduction game
Solve one case a day in five to ten minutes. There are no multiple-choice answers — connect the clues and reach your own verdict. Free and ad-supported, with no in-app purchases. Available on iOS, with Android coming next.
- iOS
- Mobile Game
- Daily Puzzle
- Free to Play
Side ProjectLive
LAST TANK
OpenAI Game Builders Seoul entry
A top-down tank PvP arena for up to 100 players in one persistent battlefield. There are no rounds — scores accumulate until you are destroyed. The real-time server runs on Cloudflare Workers with Durable Objects, kept within bandwidth budget by a 12-byte binary protocol.
- Multiplayer
- Cloudflare Workers
- Durable Objects
- WebSocket
- Canvas
Side ProjectLive
AFTERGLOW
js13kGames 2026 entry
A unicorn survivors-like that fits in a single 13 kB zip. There are no asset files — the art, sound effects, and music are all generated in code — and one build serves desktop, mobile, and gamepad alike.
- js13kGames
- 13 kB
- Procedural Art
- Canvas
- Web Audio
Side ProjectLive
Desk Smash
3D web game experiment
Type in what made you angry, then grab everything on the office desk and smash it. Three.js rendering and Rapier physics with procedurally generated 3D models and sound, running entirely in the browser in four languages.
- Three.js
- Rapier
- WebAssembly
- Procedural 3D
- 4 Languages
Side ProjectLive
GAZUA
Agentic AI trading system
An agentic AI system that rebalances a Korean stock portfolio once a day using regime-based dynamic asset allocation. A LangGraph multi-agent team — analyst panel, risk manager, PM, and a reflection reviewer — proposes each trade, while hard risk limits and a kill switch guard execution. Results are published to a public dashboard.
- LangGraph
- Multi-Agent
- Quant Trading
- FastAPI
- KIS Open API
More in the pipeline
Three products are live, all built and operated under the same model. More are already in development, and each new release is announced first at taehun.dev.
Track record
Seventeen years of shipping
Embedded Linux, then machine learning, and now a company of one. Each stage built the foundation for the next.
View full profile2008 – 2017
Embedded & Linux systems
Worked close to the hardware, building bootloaders, kernel drivers, and board support packages for network chips and single-board computers.
- Linux Kernel
- Device Drivers
- Bootloader
- BSP
- ARM64
- TCP/IP
- Virtualization
2019 – 2025
AI & ML engineering
Built production machine learning systems, from on-device SDKs and distributed training to MLOps platforms and their underlying cloud infrastructure.
- TensorFlow
- TFLite
- Kubernetes
- Kubeflow
- MLflow
- Argo CD
- AWS
- GCP
- Azure ML
- LangGraph
2026 – Present
A company of one
Founded Taehun Kim Software Lab in South Korea. Three products are live, each designed, built, and operated by one person working with AI.
- AI Orchestration
- Vibe Coding
- SaaS
- Mobile
- WebGL
- Cloudflare
- Full-Stack
Contact
Get in touch
For product questions, partnership ideas, or problems that need an engineer who can own the work end to end, email the founder directly — no queue, no Tier 1 support, no handoff.
Send an email- taehun@taehun.dev
- Location
- South Korea