The work
Pick a card. The picture is the project. The lines under it are what changed.
01 / 07
Which transformers to replace first
Utilities were ranking transformers from a spreadsheet. I rebuilt that spreadsheet exactly, then used 15–20 years of maintenance records, oil tests, and sensor history to say which units are likely to fail — and what a failure would cost.
Large California electric utilities
Which transformers to replace first
AI and data lead, with GridSure
15–20 yrs
Of real maintenance and oil-test history
In dollars
Each transformer ranked by what a failure costs
Later years
Tested on years the model never trained on
On site
The model stays inside the utility
01
Read the history
Maintenance tickets, oil tests, and the sensors already on the transformer.
02
Match their spreadsheet
Reproduce the method they trust today, within half a percent, before changing it.
03
Rank by cost
A likely failure is not enough. The list is ordered by what that failure would cost.
04
A person signs it
The write-up can only use numbers the model produced. Someone accountable approves it.
Before it counts: It has to beat their current spreadsheet on later years, not on the years it was trained on.
Maintenance historyOil testsSensor readingsRuns on site