I am Kobashi, a freelance engineer in Tokyo with 25+ years of experience. I design and build systems on AWS, put LLMs into real production use, and advise teams that need a technical counterpart who can operate in both Japanese and English.
You have a working product elsewhere and need it to run — and be accepted — in Japan. I cover the technical side: architecture that fits Japanese infrastructure and compliance expectations, integration with local systems, and communication with Japanese vendors and stakeholders.
You need a system built to standards your overseas partners expect, and an engineer who can explain the technical decisions in English. I work as the bridge between your team and the systems, vendors, or clients on the other side.
You have decided to use LLMs and now need someone who has actually shipped them. I build the environment and the system around the model — the part that decides whether an AI project reaches production or stalls at the demo.
Building production systems around ChatGPT, Claude, Bedrock, and Gemini — not experiments. My focus is the surrounding system: serverless architecture, data flow, cost control, and the operational layer that keeps an AI feature running after launch.
Typical deliverables
Designing new cloud environments, and reviewing existing ones that have grown past what they were designed for. Infrastructure as code with CDK and Terraform, so the environment is reproducible instead of hand-built.
Typical deliverables
Making an existing product work in the Japanese market on the technical level — infrastructure placement, integration with Japanese services, and working alongside Japanese development vendors as your technical counterpart.
Typical deliverables
Short-engagement advisory when you need a decision, not a project: technology selection, feasibility investigation, or a second opinion on a design. I also build with AI-assisted coding (Claude Code) and can set up that workflow for your team.
Typical deliverables
Recent engagements, summarized. Client names are withheld. The full project history is on the résumé.
Built a chat service integrating ChatGPT, Bedrock, and Gemini to assist call-center inquiry handling. Fully serverless on AWS with infrastructure entirely in CDK — an LLM project that reached production, not a demo. Team of 5, 18 months.
Stack Bedrock / ChatGPT / Gemini / Lambda / CDK / Python / ECS / Elasticsearch / API Gateway
Built a base system that provisions a full AWS backend environment for a new web app almost at the push of a button — CI/CD pipelines added dynamically and ECS / API Gateway scaled via CDK. Generating CDK from requirements with AI cut the build period substantially.
Stack CDK / Lambda / ECS / RDS / API Gateway / CodePipeline / Claude Code
For an SI vendor delivering to government offices: investigated how far a secure public-sector system can be taken on AWS, and proved it with sample implementations. Scoped to the technical feasibility question so the client could commit or walk away with evidence.
Stack Lambda / CDK / DynamoDB / Amplify / React / SES
Took over an existing large-scale real-time monitoring system after the operator changed vendors — investigated the undocumented system and continued development on it. The device count made load handling and auto-scaling the central concern.
Stack AWS IoT Core / Lambda / Go / ECS / Laravel / SQS / Serverless Framework
A single session on a specific question — a design decision, a technology choice, a feasibility check. The fastest way to find out whether working together makes sense.
A few weeks to answer one question with evidence: can this be built, at what cost, on what stack. Ends with a report and usually a running sample.
Ongoing design and implementation, typically as tech lead — owning architecture, troubleshooting, and technical research alongside your team.