花島 祐輝 | Yuki Hanajima
業務課題の発見から設計・実装・本番運用まで一気通貫で担当するエンジニア。
I turn ambiguous operational problems into production systems — from requirements through deployment and operations.
中古車輸出業向け業務管理システムの内製化。業務観察・要件定義から本番稼働まで単独で担当。
Built and shipped an internal business management system from scratch as a sole engineer.
- Stack: Go · React · TypeScript · PostgreSQL · GCP (Cloud Run / Cloud SQL / Vertex AI / Pub/Sub) · Kubernetes · Argo CD · OpenTelemetry · Grafana
- Velocity: 75 working days · 153 PRs · production cutover in ~4 months
- Impact: ¥8–15M cost avoidance vs. external development estimate · 2-day MVP during a data-leak incident
| Project | Description | Stack |
|---|---|---|
| marketplace-lakehouse-reference | Medallion Architecture data pipeline (Bronze→Silver→Gold) · 31 tests | PySpark · Delta Lake · Streamlit |
| context-smith | AI context compiler CLI — extracts task-relevant code for LLMs · v0.1.0 | Rust · BM25 · Tree-sitter |
| fluxion | Distributed WebAssembly job scheduler with DAG execution | Rust · WebAssembly · MCP |
| dataguard-rail | Data quality & lineage platform · 6 phases complete | Rust · Go · OpenTelemetry |
| stream-rail | Stream processing engine with tumbling windows & watermarks | Go · BadgerDB · NATS |
| wasm-runtime | WebAssembly binary parser · 136 tests · all sections (0–12) | Rust |
| reasoning-mesh | RAG-LLM orchestration — router + knowledge engine | Go · Qdrant · Ollama |
| oss-radar | OSS trending collector with semantic search · MINIPC cron | Python · Qdrant |
| scam-alert | OCR-based scam pattern detector · Cloud Run | Go · GCP Vision API · React |
| city-score | City livability scoring with original indicators · e-Stat API | Python · Streamlit |
| loopnest-core | BtoB Quote-to-Billing platform (M01–M14) | TypeScript · PostgreSQL |
| phone-guard | Fixed-line scam guard — ESP32 FSK decode + LINE notification on unknown numbers | Go · GCP · ESP32 |
| pachinko-analysis | Pachinko/slot prediction system with SHAP feature importance & FastAPI | Python · SHAP · FastAPI |
| doc-ingest | Document ingestion pipeline — external API docs / arxiv papers → Qdrant | Python · Qdrant |
Primary / 現職使用:
Go · TypeScript · React · Python · PostgreSQL
GCP (Cloud Run · Cloud SQL · Vertex AI · Pub/Sub · Secret Manager)
Docker · Kubernetes · Argo CD · GitHub Actions
OpenTelemetry · Prometheus · Grafana
Claude · Gemini · Vertex AI · RAG · Vector Search
R&D / 個人開発:
Rust · WebAssembly · PySpark · Delta Lake
Qdrant · BadgerDB · NATS · Terraform · k3s
Zenn (Japanese · 46 articles): zenn.dev/flipslidersand
技術記事・AI-native 開発フロー・OSS 実装記録
Dev.to (English · 4 articles): dev.to/flipslidersand
- I Thought I Used 1 Billion Tokens in Claude Code Last Week. Turns Out 97% Was Cache.
- AI-Native Development Flow: 4 Months, 153 PRs, Production
- The AI Is Fast. The Decisions Are Mine.
- Zenn: zenn.dev/flipslidersand
- Email: yukihanajob2025@gmail.com
- Open to: FDE / Applied AI Engineer / Solution Architect roles · Remote preferred · Relocation possible