For a kimono remake simulation tool, I processed uploaded photos and compressed them before sending, reducing data transfer so that processing could stay within the project's constraints. For an automated quoting system driven by AI analysis of CAD drawings, I built an application that parses drawings with AI-OCR and generates quotes tailored to each client.

Seminar One hosts seminar registration pages, and whenever a page went public, traffic spikes caused load-related errors. I ran load tests with k6 and tuned both the application and the infrastructure. I switched database reads and writes to use I/O completion ports for better efficiency in a multithreaded environment, separated the static and dynamic parts of each page so the static parts could be cached and served through a CDN to reduce load on the origin, and adjusted the autoscaling parameters to cut the number of errors during application startup. Sansan is a SaaS that has been running for over 19 years, and its core data, business-card records, numbers in the hundreds of millions, so improving existing features and building new ones both require performance-aware development. I wrote SQL with the behavior of the PostgreSQL optimizer in mind and tuned queries with pg_hint_plan, prioritizing performance from the data-structure design stage. I also worked on a microservice architecture built on a messaging system and was involved in migrating the cloud environment from Azure/C# to GCP/Go.

I worked on reducing the operating costs of Money Forward Cloud Accounting Plus. I implemented a pipeline architecture that periodically syncs binary data, previously stored in the database, to S3, cutting storage costs. I also built a dashboard to visualize development productivity and support analysis.

I developed machine-learning applications based on sparse modeling that take advantage of the structure of image and video data, including anomaly detection for continuous data such as ECGs and an annotation application for medical data. I also frequently handled CI/CD, mainly AWS permission management and deployment pipelines.

I built a player-matching service using Amazon GameLift FlexMatch. It matches players in real time based on their rank and play style and automatically creates a Discord chat room. I focused on concurrency control and performance.