AI that builds software your auditor can sign
Put AI to work under control
Sound familiar?
- AI app builders will cheerfully make you anything, and hand you a production system nobody can secure, patch or explain to an auditor.
- You want machine learning in production, and risk wants to know who approved the model.
- Every AI pilot runs in a corner. None of them reach the controls that would let them run for real.
What I do
I put machine learning and AI agents under the same controls a bank applies to people: separation of duties, change approval, an audit trail, a human who can override. For ML, that meant the first production application on public cloud at a Canadian bank, a machine learning risk application, with the compliance approvals secured. For agents, it's AITLA, a control framework I wrote: the controls sit at the action boundary, so no agent approves its own work and nothing happens off the record. I build my own software this way, and I'll show you the audit record.
What it has produced
Scotiabank: the deep learning credit card collections tool with DeepLearni.ng, covered by the Wall Street Journal. → Scotiabank GBM platform
GitHub: technical advisor on the white paper "Development Workflow for Data Scientists", on how data science teams define, enforce and automate the way their models reach production. → Selling software into banks
SigmaScope, a portfolio platform for advisors, built by AI developers under the rules that became AITLA. This site, built and run the same way. CO-ASTRA, capability-aligned budgeting software. → About
Black & McDonald: chaired the AI governance committee, ran Microsoft 365 Copilot training and workshops across the business, and ran business analysis and project management through Copilot, so the governance and the adoption arrived together. For a gene-therapy firm, an $8M AI Compute Access Fund application. → About
How we'd start
A working session on one real problem of yours, and a demonstration of an agent team building under the controls, with the audit record on the screen. Through technologyDo.