Case Studies
Case studies
Proof of problems solved. Work delivered one-stop, from requirements all the way to infrastructure.
- METI Minister's Award
METI Minister's Award winner | A B2B subscription SaaS that brought DX to the lumber-distribution industry
Successfully turned phone/fax/Excel analog trading into a single Web platform (SaaS)
METI Minister's Award | All 221 APIs with 0 missing-authorization findings, proven across 4 rounds of security audit
Inefficient phone/fax/Excel ordering and Excel-dependent inventory management. With no records left, confirmation took hours every day, and trading with new partners across companies was hard — severely hurting both efficiency and opportunity.
Python 3.11FlaskFlask-RESTfulSQLAlchemy 2.0MarshmallowAlembicPostgreSQL 16pytest+36View details An internal AI platform supporting program production at a major Japanese broadcaster (built a multi-service foundation and auth hub)
A self-built OIDC auth hub (BFF) binding 5 AI services with Google Workspace SSO, broadcast-quality speech synthesis, a caption typo-detection pipeline cross-checking OCR × speech recognition, and generative-AI compliance review — built on GCP with IaC
5 AI services unified under a single SSO | a self-built OIDC auth hub, ~30% faster caption typo detection, and broadcaster-grade multi-layered defense built with IaC
Multiple tools supporting siloed broadcast-production work (narration, caption proofreading, compliance review) with AI had to be delivered as an "always-on internal platform" — usable across the board with a single login, meeting the broadcaster's internal-control and security standards, and running heavy, expensive AI processing stably at production quality.
Next.js 16React 19TypeScript 5NextAuth.js v5PrismaTailwind CSS 4JestPlaywright+39View detailsAn AI video-localization and lip-sync platform
A GPU-inference pipeline that, from just a video upload, runs audio separation → transcription → translation → multilingual dubbing → lip-sync end to end | Earned #1 on the CrowdWorks contract ranking from the evaluation of this project
#1 on the CrowdWorks contract ranking | fully automated 8-language video localization, ~40% lower GPU cost
Doing multilingual video expansion by hand costs hundreds of thousands of yen and weeks per language for translation, dubbing recording, and mouth replacement. A production-grade AI pipeline that automates it all from just an upload was needed.
FastAPICeleryRedisPythonWhispervLLM音声クローンTTSDiffusion Lip-Sync+17View detailsA generative-AI voice chatbot
A RAG-powered generative-AI voice concierge that replaces and augments in-store service (a two-tier setup: a kiosk UI for visitors + a conversation-analytics console for operators)
A production voice-concierge AI | ~1.5-second responses, structurally eliminating wrong answers for specialized goods
Service for specialized goods had to work with limited staff and without quality differences between stores. Information was scattered across PDF, Excel, images, and video, and natural voice responses were essential — but existing chatbots stalled at PoC and couldn't deliver practical accuracy and response speed.
PythonFlaskFlask-RESTfulSQLAlchemyAlembicFlask-JWT-Extendedflask-executorGunicorn+30View detailsA real-time game-scoring app with multi-user simultaneous editing
For amateur baseball | an Expo + Next.js + Supabase monorepo | solo-built a zero-trust design pushing authorization into the DB (RLS on all 69 tables, 280 policies) and offline-resilient, idempotent concurrent input
RLS on all 69 tables, 280-policy zero-trust authorization | solo-built offline-resilient, idempotent concurrent input
Multiple people scoring the same game simultaneously, even going offline at a poor-reception ballpark, must not break. Further, what players, teams, scouts, and operators can see and do had to be strictly separated, with authorization guaranteed on the premise of not trusting the client.
React NativeExpoNext.jsTypeScriptSupabasePostgreSQLRow Level SecurityZod+17View detailsA subscription learning platform for financial literacy (built multi-channel billing, idempotent payments, and agent commissions in a Next.js 16 monorepo)
A pnpm + Turborepo monorepo (learner app / operations admin / 14 shared packages) | as a core full-stack engineer, led team development across deterministically resolving pricing from 6 ID/NFT-benefit sources, Stripe webhook idempotency/ordering/PII redaction, a bank-transfer-subscription state machine, an append-only agent-commission ledger, and type-safety discipline (banning as/any/enum + NeverError)
A Next.js 16 monorepo (3 apps, 14 packages) | 6-stream pricing resolution, idempotent Stripe payments, an append-only commission ledger, and a type-safety discipline in production
A billing foundation where users on direct sales, agents, NFT-holder benefits, and external migration mix — with different pricing, benefits, and payment methods (Stripe / bank transfer) — had to be handled on one foundation without double charges, ordering reversal, PII leakage, or double-counted commissions, and without regressing even as a multi-person team keeps changing it fast.
TypeScript 5Next.js 16React 19React Server ComponentsPrisma 7PostgreSQLMantine 8Zod+16View detailsA restaurant-matching site for foreign travelers
Tinder-style UI | multilingual | team development | clean architecture
4-language support with a Tinder-style UI to break the language barrier
Foreign travelers can't find the best restaurants due to the language barrier. Existing gourmet sites are Japanese-centric and hard to use.
GolangEchogoogle/wireNext.jsReactTypeScriptzodVercel+14View detailsLive URLA serverless payments platform in the environmental sector (full-stack development; led the payment-reliability layer)
Implemented across the 4 surfaces (customer / merchant / admin / in-store terminal — backend & frontend) plus shared foundation, CI/CD, and observability / DR / IAM | achieved 0 double charges in production via idempotency, atomic transactions, and zero-downtime migration
0 double charges in production, serverlessly | implemented across 4 backends + 4 frontends and led the payment-reliability layer (idempotency, atomic transactions, 13-phase zero-downtime migration)
Because real money, points, and carbon credits had to be handled across the customer, merchant, admin, and in-store-terminal surfaces, double charges and balance inconsistency were not permitted; the data model had to evolve without stopping the live service; and production-grade observability, resilience, and security had to be met.
PythonTypeScriptReactNext.jsViteMaterial-UIMantineTanStack Query+23View details