INDEPENDENT PRODUCT STUDIO · KOREAAvailable for freelance work

Ideas take shape.Products take life.

From the first idea to a product you can use. Independent web and AI products, designed, built and operated by one maker.

Explore the workLet's talk
FORM / FUNCTION · AI concept artwork
Projects

Built to be used.

Explore the actual screens, the decisions behind them, and where each project stands today.

Vocabit — Product screen · AI conversationProduct screen · AI conversation
Study AppLive

Vocabit

Vocabulary, spaced repetition and AI conversation in one daily learning routine.

Live service. Daily practice, review and conversation.

PHPMySQLReact
Open service
Case study
Problem
I wanted more control over review schedules than existing vocabulary apps offered, with grammar and conversation practice in the same place.
Approach
A PHP API with a repository layer over MySQL, a React front end, and a design system built for repeated daily use — dark-first, fixed type scale, no viewport-scaled text inside controls.
Now
Live at en.xenostone.org and used daily. Recent additions: study streaks with a monthly activity calendar, and selectable grammar vs. conversation AI modes.
SIGMO
  1. Strategy
  2. Risk limits
  3. Market monitor
React / FastAPI / Docker
System workflow · In development
Trading ConsoleIn design

SIGMO

One console for trading strategies, risk controls and live market monitoring.

Backend and tests in place. Interface in design.

ReactFastAPIDocker
Case study
Problem
Running automated trades across exchange screens and spreadsheets left no single place to see a strategy, its risk limits and the live position together.
Approach
A FastAPI console with JWT auth and encrypted exchange/LLM key storage. Configurable indicators (MACD, RSI, Ichimoku) sit alongside per-user prompt modules so a strategy is a versioned object rather than scattered settings.
Now
Backend and test suite in place (pytest + vitest); the console UI is still in design, so it is not yet live.
Kkomuli
  1. Camera + mic
  2. Vision LLM
  3. Voice + display
ESP32-S3 / FastAPI / Vision LLM
Prototype · System workflow
Hardware · AIPrototype

Kkomuli

A desktop companion that sees its surroundings and responds with a voice and expression.

Firmware and server verified on real hardware.

ESP32-S3FastAPIVision LLM
Case study
Problem
A voice-only assistant lacks visual context. I wanted a device that could respond to what was in front of it as well as spoken questions.
Approach
A XIAO ESP32-S3 Sense posts JPEG frames to a FastAPI server, which runs them through a Vision LLM and returns a Korean reply plus synthesised speech. The vision backend is swappable between two providers by one env var, so the device is not tied to a single vendor.
Now
Firmware and server both working on real hardware — camera, on-board PDM mic, 240×320 display and speaker verified end to end. Not a hosted service; it runs on my desk.
About

One maker. The whole product.

I plan, design and build products, then run them on my own infrastructure. Working across frontend, backend and AI means staying close to every part of the product.

Start with a real problem.

Scattered creative tools, difficult review schedules, missing visual context. Every project begins with something worth fixing.

Own the whole system.

From the interface to the API and infrastructure. Design decisions stay connected to how the product actually works.

Keep building after launch.

Self-hosting and daily use keep feedback close. Live services, work in progress and prototypes are clearly distinguished.

Self-hosted, self-operated Planning · Development · Deployment · Operations Python · PHP · TypeScript
Contact

Have a project in mind?

Tell me what you want to build, your timeline, and what matters most. We can work out the next step together.

Start a conversation
Email xenostone@naver.com
GitHub github.com/benjaminwoo1 ↗
Status Available for freelance & contract work

Product screen

Development capture. The live service may look different.