SOFTWARE STUDIO · JAPAN · UTC+9

We build software
and keep it running.

BUILD GEEKS is a software studio in Japan, run by an engineer with 19 years in production. We design, build and operate our own products across web, iOS and LLM systems. Client work gets the same treatment.

See it running →Read the code →

See the products Get in touch

01 / PRODUCTS

Things we built and run.

Our own products, designed and operated in-house. The first three come with their numbers and stack.

BJJ Techniques

BJJ Learning App

TRACTION

42
Countries with active users: 67%
Of users are outside Japan: 97.7%
Of new users complete onboarding: 44 → 368
Monthly active users, Apr → Aug 2026

In-app analytics, 90 days to September 2026.

Details →

KVS Bot

LLM × Value Investing Auto-Trading Bot

STACK

Python / FastAPI / Vercel Serverless
Anthropic Claude (Sonnet / Haiku)
J-Quants API / yfinance / Supabase

Details →

KaizenLab

Hypothesis Canvas Management Tool

WHAT IT DOES

Hypothesis canvas creation and management
Persona design and interview recording
Verification tracking and learning accumulation

Details →

Also in the catalogue

AlignA day counts only when three points align.StorytellerAI Autobiography App Clock QuizLearning App for Kids Real Estate Investment SimulatorCash Flow Analysis Tool for Property Investment Settle MapCompare Japanese municipalities by disaster risk, land price, and climate FocusnestYour switch for entering focus mode

All products →

02 / OPEN SOURCE

Code you can read.

Tools we needed for our own agent-driven development, then published.

toshipon/claude-code-security-audit-skillMITFull-stack security audit skill for Claude Code. One command sweeps the web app, the infrastructure and the Supabase backend against OWASP Top 10:2025 and CIS Benchmarks.toshipon/claude-code-security-assessment-skillMITHypothesis-driven security assessment for Web/SaaS systems. It models how the system works, asks what an attacker could achieve, and refuses to publish a finding it cannot back with evidence.toshipon/hypothesis-driven-development-skillsMITClaude Code plugin that asks for a hypothesis and a verification plan before a feature or ticket gets built. Works with GitHub Issues, JIRA, Linear or plain markdown.toshipon/cmux-agents-sidebarMITAttention-first sidebar for cmux. It treats cmux as a control plane for many agents, so you only visit the sessions that need a human.toshipon/kaizen-lab-pluginKaizenLab from inside Claude Code. Record learnings, manage canvases and run validation cycles with slash commands.

All repositories on GitHub →

03 / WRITING

Notes from the work.

English on dev.to, Japanese on Zenn.

2026-04-19If You're Building with LLMs, You Should Have Thought About Observability from Day One 2026-04-11How I Built a Full-Stack Security Audit Skill for Claude Code 2026-04-08I Have an AI Agent That Tests My Own Product Every 3 Hours 2026-03-29Why an SRE Engineer Built a Product Validation Tool — Bringing Observability Thinking to Product Development

If you want us on your project.

WHAT WE TAKE ON

AI agent adoption and PoC builds. Technical advisory and architecture review. SRE, cloud and IaC diagnosis, with hands-on fixes.

HOURS

Japan-based (JST / UTC+9). Reserved overlap hours: 8–11am JST for US Pacific afternoons, 4–7pm JST for European mornings.

HOW IT STARTS

Send a short note about the problem. We reply within a few business days.

Contact Services →