Full-stack engineer with ~8 years of experience, most of it in payments and fintech. Currently at ELESTYLE, a Japanese fintech, building the PCI-DSS v4.0-compliant payment platform elepay — 200k+ merchants across Japan, clients including SMBC, Kirin, and Coca-Cola. Deeply involved in the design and implementation of the merchant, payment, webhook, Partner API, and transaction-ledger core modules.

Systems where mistakes cost money: a bank-grade compliant debit flow (SMBC recurring-debit project, Java backend + frontend end to end), idempotency & deduplication (a Redis-backed persistent dedup window taming an MQTT reconnect storm), a webhook event-type refactor plus Merchant Keys API integration, PostgreSQL ledger modeling and query optimization (response times −40%+), fine-grained RBAC isolation, and observability — surfacing production issues from event-tracking data instead of waiting for user reports. Earlier, integrated WeChat Pay / Alipay / LianLian Pay across both B2B and B2C flows.

AI-native, not a chat box bolted on: spotted a support bottleneck and shipped a production AI support agent from 0 to 1 (RAG as the evidence layer + multi-turn dialogue + tool calling + human handoff on low confidence) — zero-cost operation, satisfaction +20%, manpower −30%. Also built flight-monitor, a multi-source monitoring agent (tool calling + MCP server + structured output + multi-source fallback + Playwright + TiDB) running long-term on Docker. Claude Code / Codex is my daily primary development environment, and I have hands-on experience bringing AI tooling into a team's workflow.

How I think about model uncertainty: in money-related paths, a model should only produce structured output from an evidence-grounded context, call tools under least-privilege control, and leave human confirmation, full audit trails, and a rollback path for anything high-risk — AI can diagnose and recommend, but it should not execute money movement on its own.

End-to-end delivery: from requirement clarification through technical design, frontend and backend implementation, testing, deployment, and production observation — I own the whole chain rather than a slice of it. Five years of team leadership (led a digital-twin visualization SaaS from 0 to 1 at DTStack into the group's largest, most stable cash-cow product), always hands-on: still writing core code and owning production incidents while leading.

Global remote & timezone: based in Tokyo (UTC+9), with a stable 4–6 hour daily overlap with CET/CEST (Tokyo 17:00–23:00 ≈ CET 09:00–15:00 / CEST 10:00–16:00), and I can start earlier when needed. Async by default — docs, issues, and code review carry the context. Native Chinese; working English for technical docs, code review, and async collaboration; business Japanese.