Vincent Luo

Operations Automation Lead · AI-native Operations

Metro Vancouver, British Columbia · luoluovol@gmail.com

Summary

Operations professional who builds the systems that run operations, by directing AI coding agents. I specified, decided and verified an order-operations platform that went into daily production on a live ERP on 28 September 2026: it reads dealers' purchase orders, keeps a person in charge of every decision, and writes orders, pick lists and shipping documents into the ERP.

Six years in operations across warehousing, inventory, purchasing, logistics, office operations and customer service, combined with building business systems with no IT support. I write the specification, set the rules and red lines, and require measured proof before anything ships.

Core abilities

• Operations: warehousing, inventory, order fulfillment, purchasing and logistics for a branch; customer-service coordination across two branches.
• Business-to-system translation: turning operating decisions into explicit rules, KPIs, forecasts and alerts; writing the procedures those rules come from.
• ERP and warehouse data: SAP Business One (HANA SQL, Service Layer reads and writes), warehouse management systems, Monday.com GraphQL; mapped an undocumented schema of 3,364 tables, 166 views and about 105,000 columns.
• AI agent orchestration: directing Claude Code agents through written work packets, with run classes by risk, model tiers, cross-model review and measured token cost.
• Verification: failing tests before fixes, positive controls, 30-day shadow runs against the ERP, and a release gate that allows "not released".
• Deployment on locked-down Windows: per-user services, scheduled-task watchdogs and signed self-updating releases with no admin rights.

Selected work

Order-operations platform · Aug 2026 – present, live since 28 Sep 2026

• Dealers' purchase orders are read by a rules engine and a language model; arithmetic (quantity × unit price = line total) settles disagreements.
• Sending saves a draft in the ERP and runs the ERP's own approval queries on it; only a clear result becomes an order. Pick lists, manifests, bills of lading, invoicing status and one merged PDF per invoice follow.
• Approval logic ran in shadow for 30 days first: 93.4% agreement, every difference traced. A 30-day back-test reproduced 99% of ERP-priced lines.
• Built for a locked-down desktop: shared-folder store, one loopback service per person, a scheduled background service, signed self-updating releases.
• 29,202 automated checks at a late-October release. Order list at 50 desks: 24.7 s → 0.70 s (p95). 546 forged requests refused, 0 written.

Purchase-order reader trained against the ERP · Oct 2026 · Whole orders read exactly right: 18% → 64% (held-out 17% → 62%) in five rounds over nine days.

Resident sync service · Aug 2026 · Two browser-extension jobs moved into one Node.js process with a 15-minute watchdog, DPAPI-sealed credentials and a start-up check that no two jobs write the same field.

Earlier tools · A nine-module warehouse dashboard, a read-only SQL explorer with 47 templates, a warehouse status watcher (Chrome MV3), a pick-list uploader, a manifest and bill-of-lading builder guarded by 658 tests. All work

Procedures · The branch operations procedure (50+ pages) and the customer-service procedure; trained the second branch on them.

Agent skills I authored

Seven Claude skills carrying 72 anti-patterns and 40 written lessons: platform change protocol, ERP and warehouse data access, locked-desktop deployment, tool-building standard, when to build an agent, rulebook update, work-board integration. How I built them

Experience

• Cover logistics, warehousing, daily operations and customer service for one branch; coordinate customer service across two branches.
• Specified and shipped the order-operations platform and the tools above, with no IT support; it went into daily production on the live ERP on 28 Sep 2026.
• Estimated effect from daily operation: about 4 hours a week saved on the inventory review, 5–15 minutes per order in customer service, about 80% of order-entry errors avoided.
• Wrote the branch's operations and customer-service procedures and trained the second branch on them.

• Led the warehouse fulfillment team, including performance reviews.
• Monitored SAP inventory and fulfillment KPIs; wrote procedures and redesigned the overseas container-tracking process.

• Ran operations end to end: vendor payments, payroll, production contracts and international purchasing.
• Built the recruiting and HR process from scratch.

• Managed the VP's calendar, presentation materials and stakeholder coordination.

Education

• B.Sc., Physiotherapy · Heilongjiang Bayi Agricultural University · 2009 – 2013
• English, post-secondary studies · British Columbia Institute of Technology (BCIT) · 2016 – 2017

Languages

Mandarin (native) · English (professional)

Technical toolbox

• AI: Claude Code, multi-agent orchestration, Claude Skills, MCP, Projects, prompt and instruction-system design
• ERP and data: SAP Business One (HANA SQL, Service Layer), warehouse management systems, Monday.com GraphQL API, Excel
• Build: JavaScript / TypeScript, Node.js, HTML / CSS, PowerShell, Chart.js, XLSX.js, pdf.js, Chrome Extension MV3, Git
• Microsoft 365: Office Scripts, Power Automate, SharePoint
• Delivery: NSIS installers, a MinGW-compiled C launcher, Windows Task Scheduler, DPAPI, signed self-updating bundles