Systems engineer for owner-run businesses
I build AI systems that streamline businesses.
I run three businesses and a household on 65 of them. Every one reports to a human. I spent fifteen years in finance and operations before I built the first one, so the process comes first and the AI comes second.
What I write about
Four parts of a business, one method.
Each piece takes one system apart: the problem it solved, the workflow as a diagram, what AI built for each component, where a person still decides, what it replaced, and what it costs to run.
Finance
The close, the consolidation, the loan book
Multi-entity consolidation read straight from the ledger, bank statements filed into a loan book with a review queue, a nightly job that learned to fail loudly. Written under a strict rule: no entity, vendor or ERP named.
About a third of the pieces
Real estate
Lists, outreach and underwriting for a brokerage
Owner lists built from public records, lead pipelines that draft and never send, distressed listings underwritten every night, a market pull with no model in the data path. Shown with the brokerage named.
About a third
Operations
How one person runs the rest
A chief of staff that briefs me at 06:40 for 54 cents, a kanban board where the agent does the building, a to-do list reviewed before I wake, meeting notes that file themselves.
One in five
Household
The same method at home
A first grader's reading measured against a published benchmark, a household close that costs 46 cents a quarter, meals planned from what actually got cooked.
One in six
How I work
Five rules every system here follows
Every system reports to a person
Nothing sends, pays, deletes or signs on its own. The system drafts, flags or proposes; a human taps yes or no. Autonomy is earned per task type, after a measured run of good proposals.
Process first, then the machine
I map the workflow before I automate a step of it. Most of the time the fix is a missing control or a duplicated step, and the AI part is small.
Read-only until proven
A new system reads for weeks before it may write. It is rehearsed on a full copy of last month's data, then tied out against the signed figures.
Green is not correct
A job that reports OK can still drop two thirds of its rows. Every pipeline checks its own counts against a second source and stops when they disagree.
Know the run cost
Every system has a cost per run and a monthly cap. A morning brief is 54 cents. A quarterly household review is 46 cents. A comp search is capped at two dollars.
About
I started as a commodity trader in Singapore, learned trade finance from the inside, and earned the CAIA charter to understand how capital is priced. I raised a Series A to build trade finance software and moved to California to bring it to the Americas.
Today I run AI inside a multi-entity trading group's finance and operations, co-own a property brokerage and a produce export firm in Northern California, and raise two children. None of those came with an IT department, so I built the systems myself. The finance work appears here with names, vendors and systems removed; the rest appears as it is.
I work with owners who want the same thing built for their business: advice first, then the build. Write to [email protected].
One email a fortnight
One system, taken apart, with the figures.
The problem, the workflow, what AI built for each part, where a person still decides, and what it costs to run. Nothing else, and nothing for sale.
One line is enough. I add you by hand and reply within two working days.