Privacy
Patterns, not private artifacts
Public examples are scrubbed of real names, vendors, and confidential figures. The method can be inspectable without exposing the people inside the work.
The person behind the checks
About Ali
Afsar Ali, known as Ali, is a data scientist who tests AI systems on real work — then publishes the evidence needed to decide whether the output deserves trust.
I’ve worked for more than a decade across forecasting, attribution, and marketing analytics, including roles at PACCAR, Zillow, and a large technology company.
I use AI heavily. I also assume that a fluent answer can be wrong, incomplete, stale, or impossible to reproduce. The useful part is not the demo; it is the check that catches that failure before a person relies on it.
How I work
Operate, challenge, publish
01
Use it for real
Test the system inside an actual workflow where missing or stale data has consequences.
02
Challenge the output
Reconcile sources, apply deterministic checks, and ask an independent pass to make the claim fail.
03
Publish the boundary
Show what held up and what the evidence still cannot establish.
Accountability
Corrections stay visible
When a claim changes, the correction is labeled and dated instead of silently overwritten.