Methodology
My science is my method.
My doctoral research was, in plain terms, the design of systems that separate one thing from another with precision — and proving, under harsh conditions, exactly how well they do it. The discipline transferred directly.
- 01
Define what “working” means — from the problem, not the tool.
In the lab, I set the target from the application — the exact separation a power plant needs — before choosing any material. In AI, success is defined from your objective, not a vendor’s benchmark.
- 02
Measure against reality — and never trust a single number.
A strong headline metric can still hide a defective system, so I judged quality with a second, independent signal. In AI, no system is certified on one benchmark.
- 03
Tell the real result from the artifact.
Much of the work was tracing a surprising reading back to contamination in the apparatus, not reporting it as a discovery. In AI, that is separating genuine failures from harness bugs and flukes.
- 04
Report what you can’t yet explain.
When data and theory disagreed, I reported the gap rather than hide it. That habit — measuring honestly, saying what I don’t yet know — is built into every system I design.
The same method, applied now to AI. You can see it at work in the Teardowns.