Daeseon Yoo
Daeseon Yoo avatar

Daeseon Yoo

Backend & AI Engineer · Toronto, Canada

Open to roles · Remote or Toronto

Software engineer with 6 years modernizing complex production systems across manufacturing, warehouse, and financial operations, and shipping measurable results: processing 60% faster, monthly reprocessing cut to near zero, a 7,700-line core query reduced 57%. Deep in API and data-layer design, transaction integrity, and concurrency across distributed multi-server systems — in Java/Spring, C#/.NET, and SQL/PL-SQL. Now in Toronto, building Spring services and LLM-powered developer tools.

Currently

  • Recently wrapped up ~5 years at SK AX (Korea, Sep 2021 – May 2026). Now based in Toronto.
  • Building LLM-powered developer tools — TubeShadow and Dalkkak — and this site, under the Daeseon AI Factory umbrella.
  • Open to senior backend or AI engineer roles. Toronto or remote.
  • What I'm on right now lives at /now.

Previously

  • SK AX · Software Engineer · Korea

    Sep 2021 – May 2026

    Manufacturing cost management, an enterprise mobile web platform across 8 sites, and real-time MES. Backend Java/Spring, JPA, PL/SQL on Oracle, and transaction redesign across manufacturing and financial systems.

  • Dure Info · Software Developer · Korea

    Jun 2020 – Sep 2021

    Multi-client TCP socket server (Windows Service) brokering DB access for PC/PDA/Kiosk clients. Fixed XML deserialization failures from partial TCP reads with application-level message framing — buffering bytes until a complete delimiter-terminated message arrived.

Selected impact

  • Consolidated 40 endpoint-specific middleware APIs into one reverse proxy (reflection-based RPC dispatcher), deployed across 8 sites — freeing ~1 FTE by eliminating per-endpoint redeployments.
  • Redesigned transaction boundaries between manufacturing and financial systems with a staging-table + batch-worker pattern. Processing time −60%; reprocessing from 20 cases/month to near zero.
  • Refactored a 7,700-line monolithic CASE-based PL/SQL query into per-process CTEs (−57%), isolating 5 sub-processes so changes stop cascading. Weekly cost-ledger throughput 4 → 7 tickets.
  • Fixed HTTP 413 errors on mobile WMS scans with PK-only payloads + fresh-state retrieval — no infra change. Scan batch size 30 → 60, inventory stayed accurate.
  • Eliminated cross-server duplicate execution with a distributed lock (ShedLock / job_lock table) and decoupled scheduler from execution via @Async, so parallel plant closes run safely.
  • Prevented duplicate ID generation across multi-instance MES with Oracle row-level locks (FOR UPDATE), avoiding data anomalies and downtime.

Writing about

  • LLM app patterns — async analysis pipelines, provider abstraction, prompt engineering.
  • Backend systems — transaction design, locking, concurrency, cross-system integration.
  • Project recaps in the before / change / result / limit format.

Stack

Languages
Java, Python, SQL/PL-SQL, C#, TypeScript
Backend
Spring Boot, JPA, FastAPI, REST APIs, .NET
Frontend
Next.js, React
AI / LLM
Claude & Gemini API, prompt engineering, async LLM pipeline
Databases
PostgreSQL, Oracle
Infra
Docker, Git, Linux, AWS, CI/CD (GitHub Actions), JUnit

Education

M.S. & B.S. in Industrial Engineering · Kumoh National Institute of Technology, Gumi, South Korea (2012–2019).

Get in touch

Email is fastest: showep12@gmail.com. Résumé: /resume.pdf.