Wallace Lo · Ontario, Canada
Model-risk tooling and evaluation harnesses for Canadian financial services ahead of OSFI Guideline E-23 [1↗]. Insurance professional turned builder · CIP 7 of 10 [2] · systems graded against realized outcomes [3]. What this page can't yet cite, it says so: the first public case study .
I spent a decade inside Canadian P&C insurance [4↗] watching model decisions move real money and real claims. Now I build the tooling I wished the second line had · through flowright, the agency I run [5↗] and in the open.
The through-line is provability. A system that can show its evidence gets trusted with more than one that just performs well · and the regulated industry I come from is about to require exactly that.
Certified AI governance professional
AIGP · exam 2026-09-14
An options-research tool that grades its own past calls [6] against market-adjusted outcomes · an honest calibration curve, not a profitable-bot claim. Nothing auto-executes.
fig. 1 · session replay · last 10 calls graded, 7 correct · drawn by the shipping painter, x-rayed under your pointer
agency · locks point outward
My agency's platform takes regulated documents from intake to audit-ready evidence · every output cited to its source or abstained. DocIntake [7↗] and regRAG [8↗] are its proven components; the product depth lives at flowright.io [9↗].
fig. 2 · the provenance loom · 8 claims: 6 cited, 2 abstained, 1 repaired · a live window onto the shipping figure
Claims about unshipped work don't get citations here. They reach, in ember, and hang until they're true.
the DocIntake engine, repointed at model evidence · unshipped
running now · becomes claim [4] when it closes
real footage, one honest failure story each · recording