I’m a cloud architect and AI engineer focused on what lies beneath the interface: resilient infrastructure, clear orchestration, and software that holds up under real-world load. At IBM Japan, I architected multi-region AWS banking platforms with 15-minute recovery objectives and led engineering teams from system design through production cutover.
Now based in California and studying Data Science at UCLA, I build open-source, local-first AI agents, multi-CLI orchestrators, and end-to-end LLM fine-tuning pipelines — pairing enterprise cloud discipline with hands-on AI engineering in English and Japanese.
Cloud
Architecture on AWS, Azure, and Google Cloud, with a focus on high availability and multi-region disaster recovery. Hands-on with RDS, Lambda, API Gateway, CloudWatch, and Amazon Connect. Recognized as a 2025 Japan All AWS Certifications Engineer.
AI
Local-first agents and LLM applications built with Claude, Codex, and Amazon Bedrock, from orchestration and evaluation to harness design. Hands-on LoRA fine-tuning with PyTorch, PEFT, llama.cpp, and Ollama.
Full stack
Java and Spring Boot, Python, C#, and TypeScript, with Node.js, React, and Next.js. Backend services, integrations, and CI/CD, delivered from requirements through production.
Beyond the screen
Open roads on a motorcycle, unhurried travel, golf, and a quiet pour-over at a café I haven’t visited yet — places where the air clears and the next idea takes shape.
Selected projects.
Aitne
Local-First Personal AI Agent
A personal AI agent that runs around the clock on your own machine. A long-running TypeScript daemon uses Claude Code, Codex, or Gemini CLI as its brain: it plans the day each morning, watches mail, calendar, and GitHub, keeps its memory in plain Markdown, and reaches you through Slack, Telegram, or Discord. Open source and published on npm.
Runs local AI command-line tools — Claude, Codex, and Gemini — from Slack, with a web dashboard for sessions and workflows. A dependency-aware DAG engine coordinates parallel tasks, webhooks, and schedules, while secrets stay in the OS keychain. Its architecture now carries forward into Aitne.
An end-to-end fine-tuning pipeline adapting Google’s Gemma 3 (1B and 4B) for domain-specific dialogue. Built on Apple Silicon, the workflow spans dataset curation and PEFT/LoRA training through adapter merging, GGUF quantization, and local inference on Ollama — systematically evaluating chat templates, stop-token boundaries, and precision trade-offs.
Studying Data Science to deepen the statistical modeling, quantitative evaluation, and machine learning foundations behind the AI systems I build.
2023 — 2025
IBM Japan
IT Specialist & Cloud Architect · Backend engineering & team leadership
Architected and delivered mission-critical banking platforms on AWS across the full lifecycle, from requirements and disaster-recovery design through production cutover. Progressed from sub-lead to acting team leader for teams of up to six engineers.
Released / October 2024
SWIFT Payments Platform
Serving as IT Specialist and Cloud Architect (lead engineer for a six-person sub-team within a 30-person program), drove the application and AWS modernization for the ISO 20022 migration. Architected multi-region warm-standby disaster recovery with a 15-minute RTO, owned backend delivery from design to production, and kept critical payment services running seamlessly through cutover.
AWS / RDS
Java / Spring Boot
C# / Python
CI/CD
Nginx
PowerShell / Bash
Released / August 2025
Contact Center Platform
Serving as IT Specialist and Cloud Engineer (tech lead and acting team leader for a four-engineer team), architected and built a cloud contact center platform for a major Japanese bank. Integrated Amazon Connect with Salesforce Service Cloud Voice through AWS Lambda and API Gateway with end-to-end CloudWatch observability, leading delivery from requirements through production readiness.
Amazon Connect
Salesforce
Lambda / API Gateway
CloudWatch
2019 — 2023
Hosei University
Bachelor of Economics
Studied microeconomics, game theory, and behavioral economics — how people make decisions, and how the systems around them shape those choices.
Loop Engineering replaces you — the person prompting the agent — with a system you design. The four layers that led here, what a loop actually is, the principles that keep one honest, and a worked implementation you can run today.
The evaluator — the mechanism that grades how well an agent performs — is the most postponed part of the harness, yet the part that decides whether you reach production and can ride the coming self-improvement wave. Why it belongs at the start of design, not after deployment.
Putting a rule once in CLAUDE.md vs. re-pasting it every turn — measured against the real claude --print. The fixed cost of framing, the re-send penalty multiplier, and how conflict resolution splits by model (Sonnet stays faithful, Opus leans to recency).
The meta-harness skeleton survives the move from an individual user to an enterprise — but several design assumptions flip. A two-layer scope design for self-improvement loops you can actually operate.
What happens when you build a personal AI that can observe and modify its own harness — and how to implement it without letting it spiral out of control.
Let’s build something good.
I’m pursuing full-time AI, Cloud, and Forward Deployed Engineering roles in the U.S. — ideally in San Francisco or Los Angeles, and open to strong teams nationwide or remote. Through my company in Japan, I also take on select cloud architecture and applied AI consulting engagements in English or Japanese.
