HAMADA Taigen

• Shipped LLM-based document extraction to production at a healthcare SaaS — replaced a legacy OCR pipeline with substantially higher accuracy. FastAPI + LangChain v1/LangGraph multi-agent orchestration with multi-model support (Claude, GPT, Gemini, Mistral), async task-queue workers, and HMAC-signed webhooks on AWS.
• Built the LLM evaluation stack: golden-dataset experiments on Langfuse (observation/user/session-level scores, trace costs) with dedicated CLI tooling, plus shadow deployments comparing candidate models on live traffic before switching defaults.
• Release engineering: rebuilt container CI/CD around a build-once model — prebuilt base images plus an ECR registry cache cut image builds from 17 minutes to under 5; added automated rollback and reusable deploy workflows (QA/staging/prod).
• Test infrastructure: led a 9-part RSpec optimization project on a 100k+-commit Rails/Vue monorepo (CircleCI Gen2, test-suite splitting, auto-rerun) — individual suite runtimes cut by up to ~40%.
• Daily production user of Claude Code, Cursor, GitHub Copilot, Codex, and Devin — reviews and corrects AI-generated code as part of normal work; distilled the know-how into agent skills (LangChain, Langfuse, FastAPI, Terraform, optimize-rspec) and custom sub-agents shared across the engineering team.
• Spec-driven development: write thorough design docs and context so coding agents deliver the intended result on the first pass.

AI Tooling ★

Claude Code · Cursor · GitHub Copilot · Codex · Devin · LangChain · LangGraph · Langfuse

Languages

TypeScript · JavaScript (ES6) · Ruby · Python · PHP · Java · + Kotlin, C++11, Zig, Lua, Rust, Go, Elixir, Nim, Scala

Frontend

React · Vue.js · Next.js · Gatsby · Astro · TanStack Start · Solid Start · React Router v7 · React Admin · Svelte · Storybook · jQuery · Sass

Backend

Ruby on Rails · Node.js / Express · FastAPI · Django · Spring Boot · CakePHP · FuelPHP · Ktor · Phoenix

Databases

PostgreSQL · MySQL / MariaDB · MongoDB · libSQL

Cloud / Infra

AWS · Google Cloud (Cloud Run) · Docker · Terraform · GitHub Actions · CircleCI · Datadog · Firebase · Supabase · Cloudflare

Auth / Others

Auth0 · Clerk · Twilio · Vite · Webpack · Git

★ = daily production use

1. iCARE, Inc. — Carely (healthcare SaaS)

   Full-Stack Engineer / Tech Lead (PL/PM) · Nov 2022 – Present · Tokyo, Japan

   Vue.js, Ruby on Rails, Python, FastAPI, LangChain/LangGraph, AWS, PostgreSQL

   • 4,574 commits / 851 merged PRs across the main SaaS and the AI-OCR platform — one of the primary engineers on the LLM document-recognition service (976 commits / 116 merged PRs).
   • Built the production LLM pipeline (Jan 2025 – Apr 2026): FastAPI + LangChain v1 / LangGraph multi-agent orchestration, multi-model support (Claude, GPT, Gemini, Mistral), structured extraction of exam items/values/units/judgments, async task queue (Taskiq + PostgreSQL), HMAC-signed webhooks, S3/MinIO storage, JWT auth.
   • Built the LLM evaluation stack: golden-dataset experiments on Langfuse (observation/user/session-level scores, trace costs) with dedicated CLI tooling; shadow deployments comparing candidate models on live traffic before switching defaults.
   • Led a dedicated AI OCR R&D phase (Oct 2024 – Oct 2025): evaluated models and approaches on real-world documents, then designed and shipped the production pipeline.
   • Health-checkup features on the main SaaS (300+ commits): LLM-based ingestion of scanned checkup results; designed a configurable rules engine for health conditions (e.g., metabolic syndrome criteria) that later features reused.
   • Vue 2 → Vue 3 migration: eliminated hundreds of typing errors across shared components, fixtures, and GraphQL codegen layers (single PRs removing 273 / 83 / 42 errors).
   • PII data masking for staging environments: full masking-policy audit, RFC 2606 reserved domains, credential redaction, outbound-mail suppression.
   • Release engineering and observability for the OCR service: build-once CI/CD (image builds 17 min → under 5 min), automated rollback workflows, Datadog APM dashboards, and retry instrumentation for queue workers.
   • Developer platform (2026): led a 9-part RSpec optimization of the Carely monorepo — CircleCI Gen2, test-suite splitting, auto-rerun, suite runtimes down up to ~40%; migrated frontend linting to type-aware oxlint and upgraded Bundler to 4.
   • Engineering enablement: Claude Code workflows, Agent Skills and custom sub-agents, CLAUDE.md coding rules, Cursor Cloud Agent environments, and CI merge gates — rolled out to the entire engineering team.

2. Legalscape (legal-tech SaaS)

   Independent Contractor — 24 hrs/week · Jun 2024 – Jun 2026 · Remote, Japan

   Vue.js, React, React Admin, React Router v7, Node.js, Express, TypeORM, Python

   • Delivered new features and a frontend redesign while working 24 hrs/week in parallel with full-time employment — a proven part-time remote contractor track record.
   • Full-stack ownership on a 3-person team: frontend and server-side development.

3. FAIR NEXT INNOVATION

   Full-Stack Engineer / Tech Lead (PL/PM) · May 2019 – Oct 2022 · Tokyo, Japan

   Next.js, React, Gatsby, Node.js, Ruby on Rails, Kotlin/Ktor, GCP (Cloud Run), Auth0, Twilio

   • Led 5 web services from inception through design and launch as PL/PM and hands-on engineer on teams of 3–6.
   • Public gym web reservation service (Saga Prefecture): Gatsby + Rails + GCP Cloud Run + Auth0 — designed the entire admin UI/UX myself where no designer was assigned.
   • Real-time video streaming service (Next.js + Kotlin/Ktor); web gacha and web voting services (Next.js + Rails); contact-center internal chat tool (Next.js + Twilio) shipped on a tight schedule with zero critical bugs.

4. Silicon Studio

   C++ Engineer · May 2018 – Apr 2019 · Tokyo, Japan

   C++11, Qt, Visual Studio, in-house rendering engine (Mizuchi)

   • Built a digital dashboard mock tool (Qt) used to prototype next-generation vehicle instrument clusters for a major automaker.
   • Improved driving-simulator graphics: reflected vehicle telemetry (speed, steering angle) in real-time scenes rendered by the in-house engine.

5. Network21 / Digital Hearts

   Software Engineer · May 2016 – Apr 2018 · Tokyo, Japan

   Java, Spring, Objective-C, Python, Scrapy, Selenium

   • Developed an iPad/iPhone POS register application: Java/Spring backend + Objective-C iOS app.
   • As main programmer, built a web scraping tool that detects cross-site-request vulnerabilities — launched as a new business line with significant early revenue.
   • Continuously developed test automation (Java/Selenium) on a 5-person team for Digital Hearts, the group's parent test-services company (2017–2018).

1. gitz

   A terminal git TUI in Zig, aiming for JetBrains-grade interaction

   A commit-focused git TUI with a three-pane Changes/Diff/Commit flow, plus a log-view context menu for revert, cherry-pick, reset, tag, branch, and compare operations. 380+ commits of active solo development.

   Zig 0.16 · zigzag

2. Mortgage Loan Simulator

   Japan-market mortgage simulator: variable-rate scenarios and the 2026 tax-deduction rules

   Simulates variable-rate repayment scenarios, the 2026 mortgage-interest tax deduction, and early-repayment planning for the Japanese market. TanStack Start + React 19, Tailwind v4, shadcn/ui, Recharts, Zod, TanStack Form, Cloudflare Workers.

   TanStack Start · React 19 · Cloudflare Workers

3. Veritra

   Generates test cases from requirements and viewpoints

   A test-design web app that generates test cases from requirements and analysis viewpoints — viewpoints defined per feature become reusable test-design assets. Node.js 22 / pnpm / libSQL / Cloudflare Workers, with a test suite and CI lint.

   Node.js 22 · libSQL · Cloudflare Workers

4. jobcan-chrome-ext

   Chrome extension improving a time-tracking system's UX

   Chrome extension improving the UX of a time-tracking system's work-hour entry screen: reload prevention, month persistence, and continuous batch input.

   Chrome Extension

5. codex-plugin-cc

   Call Codex from Claude Code for code review and task delegation

   Fork/customization of AI-agent tooling: invoke Codex from Claude Code to review code or delegate tasks — a second-opinion reviewer inside the coding-agent workflow.

   Claude Code · Codex

6. cline-kanban

   Run multiple CLI coding agents in parallel from a local web app

   Fork/customization of AI-agent tooling: launch a local kanban web app and run CLI coding agents in parallel — orchestrating multiple agents across tasks.

   CLI coding agents · Web

Open to contract work, AI evaluation projects, or a quick tech chat: