Five years of general ledger and ERP work at GE — Oracle migrations, ASC 606 revenue recognition, $200M+ renewables projects. Now I build the automation I used to file tickets for.
Five years owning general ledgers across Australia, Mongolia, Thailand and Vietnam. Led an Oracle CCL to RACES migration end to end — data mapping, validation, cleansing, UAT. Closed 50+ reconciliation accounts with zero unreconciled balances straight through a system transition.
At GE, automating cost accrual and revenue recognition cut processing time by 75%. That instinct did not switch off when I left. I now write the software myself — Python for finance pipelines, TypeScript and React for interfaces, SQLite for local-first apps.
Every project here ships with a written spec, a phased plan, and conventions the model has to follow. The interesting part is not that AI writes code — it is what has to be true about your thinking before it writes the right code.
2025 — ongoing
Ingests CSV exports from eight banks and card issuers, normalises schemas that agree on nothing, categorises every transaction against a rules engine, and produces quarterly spend reports. The problem I solved at GE, rebuilt solo.
2026
A local-first climbing and fitness journal for iOS. No accounts, no server — SQLite on device, HealthKit for the numbers it should not make you type twice.
2025
Time accounting in the tradition of Alexander Lyubishchev, who logged every hour of his working life for 56 years. Built to find out where my days actually go.
The workflow I use with AI — a spec, a conventions file, and a handoff note — and why deciding what the model may not do turned out to matter more than describing what it should.