I build AI-powered automation tools that solve real business problems.
Mechanical Engineering graduate. Working in project controls, building production software on the side. Founder of NUMP Limited — a registered NZ company and member of AI Forum New Zealand.
Production mobile app for NZ shift workers to verify their pay — before it hits their bank account.
NZ shift workers are paid under rules most don't fully understand: PAYE is calculated on annualised income, not a flat rate. Penal rates apply per-minute across shift boundaries. Mondayization changes the gross amount. ShiftPay gives workers the single source of truth — log a shift, see exactly what you should earn, before the payslip arrives.
Full payroll engine: PAYE, ACC, KiwiSaver, student loan, penal rates, Mondayization. The finance engine processes each minute of a shift independently — the only approach that correctly handles shifts crossing midnight into a public holiday. RevenueCat subscription management, biometric authentication, offline-first via Drift/SQLite. Solo built, end-to-end product ownership.
App Store · Google Play · Case Study · Repo
Privacy-first AI tool that reads NZ payslips, recalculates every deduction from scratch, and tells workers whether their employer paid them correctly — without storing a single byte of their data.
Upload a payslip (PDF or image), get a line-by-line audit in seconds. No account. No data stored. No friction. Architecture was designed so there is nothing to collect: the payslip never touches a database, the result is deleted the moment it's read, and privacy compliance is enforced by the code — not a policy document.
The core AI challenge wasn't reading payslips — it was getting structured, machine-readable output every time across hundreds of different NZ employer formats. Solved with Claude's forced tool use (tool_choice: { type: "tool" }) and a typed output schema covering every NZ payroll component. Background function architecture handles the 15–30s Sonnet response time within Netlify's free tier.
AI-native business operating system. Claude as the intelligence layer — tool use, RAG pipeline, multi-tenant agent backend.
GCP Cloud Run proxy, Supabase + pgvector, Vertex AI, Claude API. The ANTHROPIC_API_KEY never reaches the browser — all credentials managed server-side via GCP Secret Manager. CI/CD via GitHub Actions + GCP Workload Identity Federation (keyless auth). Production infrastructure, not a prototype.
Automated workplace safety system — identified the operational gap, designed the solution, deployed it.
Multi-tier escalation: user notification → manager alert → Managing Partner if check-in missed. Integrates Power Automate, SharePoint Lists, Teams Adaptive Cards, and Google Maps API for real-time location verification.
| Area | Technologies |
|---|---|
| Mobile | Flutter · Dart (two live stores) |
| AI / LLM | Claude API · Tool use · RAG · Multi-tenant agent architecture · pgvector |
| Backend / Cloud | Supabase · GCP (Cloud Run, Firebase, Vertex AI) |
| Enterprise Automation | Power Automate · CoPilot Studio · Power Apps · Power BI |
| Integration | REST APIs · OAuth · Webhooks · Custom connectors (BambooHR, Google Maps) |
| Frontend | JavaScript · Tailwind CSS |
| DevOps | GitHub Actions · GCP Workload Identity Federation · CI/CD |
- BBR agent backend — pgvector RAG pipeline + Claude tool use
- Learning Python / FastAPI — targeting production comfort by mid-2026
- TypeScript being added to BBR frontend
- GCP Associate Cloud Engineer certification in progress
Mechanical Engineering degree. I work in project controls at a project management firm — which means I understand how organisations actually manage risk, cost, and schedule before writing a line of code to solve their problems. That context is what separates the Travel Register (deployed, used daily) from a demo.
Founder of NUMP Limited — a registered NZ company and member of AI Forum New Zealand.
NUMP Limited · LinkedIn · ShiftPay — App Store · ShiftPay — Google Play · robert@shiftpay.co.nz