Senior Full-Stack Engineer
Autify, Inc. · Tokyo, Japan
April 2022 – Present
Test automation SaaS, 300+ customer companies. Full-stack across its three products and the team's internal platform.
AI-powered web app testing SaaS — Built Rails full-stack features and Chrome, Edge, Firefox and Safari automation
- Retired an ML GPU cluster by replacing it with a deterministic DOM-based resolver, shadow-tested in production behind a canary flag to 99.5% agreement before rollout
- Upgraded the Kubernetes (EKS) clusters four minor versions and the Karpenter autoscaler from 0.x to v1; right-sized Windows Edge standby pools from a month of real demand, cutting standby by about a quarter
- Kept browser automation working through regular browser updates (Chrome/Edge 124 → 152, Firefox 125 → 155), debugging native Selenium WebDriver and Chromium internals
- Built Chrome extensions for recording and replaying browser actions (TypeScript, React, Chrome Extension API)
Native mobile app testing SaaS — Built Rails full-stack features for iOS and Android device automation
- Reduced mobile device wait time by ~50% by pre-warming devices before test execution, and drove the frontend's migration from a Rails multi-page app to a React / TypeScript single-page app
Desktop-based web app testing SaaS — Built TypeScript full-stack features for natural-language test generation
- Deployed a self-serve infrastructure provisioning API (Pulumi) with AWS IAM permission isolation
- Consolidated the desktop application's multi-region deployment onto a single cell end to end: built the per-organization maintenance mode and rollback the migration used, then removed the retired cell's deploys, routing and infrastructure
- Drove development of the scheduling feature (Electron, TypeScript, React), and made the PDF export safe to interrupt
- Prototyped a natural-language AI agent (OpenAI) for mobile test automation, accessed through the desktop application: prompt adherence, element location, model selection
Built AI tooling, PR review and dependency workflows, and AWS infrastructure
- Built a self-hosted dependency-update platform (Renovate as a GitHub App) that consolidates security alerts and CVE reports, merges minor patches quickly and gives the rest an automated fix attempt checked against CI; rolled out to 16 repositories
- Built an AI PR reviewer on GitHub Actions with Claude, cutting cost per review from $2 to $0.20; replays on past PRs picked the model and kept it quiet on most PRs human reviewers had passed
- Sandboxed the dependency pipeline's AI agents after finding that one write token could reach every managed repository: per-repo least-privilege App tokens, assess and fix in separate locked-down non-root containers, and the signing key behind a protected environment
- Built MCP servers in Ruby for the web and mobile products, and an incident-triage bot over 7 MCP servers under read-only IAM policies
- Created an ETL pipeline (AWS Glue / Python) that processed 20M+ unstructured log records in days instead of weeks
- Drove data analytics initiatives across the Business Intelligence stack (Holistics, Redshift, Athena, Airbyte): cost tracking and data governance
- Architected email delivery (SendGrid), and mentored QA and technical support engineers on making code contributions
- Took part in on-call incident response for dozens of incidents since mid-2024 alongside a few dozen engineers across SRE and product teams, acting as incident commander on a couple of major ones
Full detail
- Built web product features primarily on the Ruby on Rails backend (GraphQL, PostgreSQL, RSpec), and dogfooded the product for QA
- Experimented with the Devin AI agent for automated dependency updates
- Built an AI pull-request reviewer on GitHub Actions with Claude, cutting cost per review from $2 to $0.20: structured findings posted as inline comments, threads auto-closed once a later push fixes them, and per-run budget and time caps; replays on past PRs picked the model and kept it quiet on most PRs human reviewers had passed, and a two-agent self-review pair caught several times as many real issues as a commercial tool
- Built MCP servers in Ruby for the web and mobile products, with safeguards against rate limits and system load, and batched responses to keep token consumption down
- Built an incident-triage bot on an in-house agent runner that connects to 7 MCP servers gated by strict read-only AWS IAM policies, iterated continuously to improve its signal-to-noise ratio
- Upgraded the NoCode Web staging and production Kubernetes (EKS) clusters four minor versions and the Karpenter node autoscaler across a major version (0.x to v1), stepping one version per change and resizing its controller requests for v1
- Right-sized standby pools of Windows Edge test machines from a month of real demand, cutting standby by about a quarter; projected to save hundreds of dollars a month, and a replay of a month's traffic took wait-limit timeouts in the busiest pool to zero
- Maintained the test automation pipeline for web / mobile automation, serving 300+ companies
- Architected email delivery (SendGrid); prototyped server-side PDF generation (wicked_pdf) and contributed the print layout for the browser-based PDF export
- Made the desktop application's PDF export safe to interrupt and let PDFs download instead of opening a viewer
- Kept browser automation working through regular browser updates (Chrome/Edge 124 → 152, Firefox 125 → 155), fixing the breakage they caused: Edge self-updating mid-run and crashing its driver and a Chrome build missing on Linux
- Assisted in building the test automation pipeline for a desktop application; drove development of the scheduling feature (Electron, TypeScript, React)
- TypeScript
- React
- Apollo GraphQL
- Electron
- Chrome Extension API
- Appium
- Ruby on Rails
- GraphQL
- Postgres
- SQLite
- Github Actions
- CircleCI
- Jest
- Testing Library
- Mock Service Worker
- RSpec
- Playwright
- AWS (ECS/EKS)
- Kubernetes
- Pulumi
- Terraform
- Holistics
- Athena
- Redshift
- Airbyte