Tokyo, Japan. Oversee product engineering and lead a team of 6 backend engineers and 6 frontend engineers building LLM AI agents for the architecture, engineering, and construction industry. Built a successful MVP from the ground up; shipped multiple PoCs and reworked core code.
- Owned the company's technical direction — set architecture standards, technology choices, and engineering roadmap.
- Grew engineering from 4 to 12, hiring 8; manage all 12 directly (6 backend + 6 frontend); own hiring and performance reviews.
- Own a ~590K LOC production codebase held to a 40% unit-test share, spanning frontend, backend, microservices, queues, workers, and schedulers; main ships a production release every week.
- Built the MVP from zero to first enterprise deployment in ~6 months.
- Designed and architected systems; built the data layer, API layer, FE layer, and AI agent for AI and LLM consumption; facilitated Elasticsearch integration.
- Implemented async / coroutine‑based task runners and a real‑time application layer.
- Designed and implemented real‑time events using SSE and ElasticSQL to power streaming chat, live dashboards, and notifications.
- Designed a WebSocket + RabbitMQ streaming solution implemented as a FastAPI microservice as an option for real‑time event notification.
- Built a data ingestion pipeline — PDF → OCR → structured data — on RabbitMQ and FastStream processing ~50K pages/day of PDFs and 3D models, cutting manual extraction from hours to minutes per document set.
- Built an agentic AI harness for architectural data — retrieval, manipulation, and modification through LLM‑driven tool calls.
- Routed across 3 LLM providers (OpenAI, Anthropic, Gemini) via Pydantic AI, cutting cost per request ~30%.
- Rolled out AI coding agents to all 12 engineers; merged PRs per engineer up ~2x.
- Filled gaps as PM and DevOps when short‑staffed; oversaw and ran engineering interviews; researched bleeding‑edge AI tooling.
- Ensured code quality across the team via review, mentorship, and architecture standards.