Auditable production · Regulated industries
If you can’t audit an AI system, you don’t control it.
I design and lead AI systems that stay auditable in production, in regulated industries.
Vaibhav Khandelwal
AI architecture and delivery · PwC
I lead AI architecture and delivery for regulated-industry clients at PwC. Views are my own.
Three questions before an AI system goes live
What it may do. What evidence shows what it did. Who owns it when behavior changes.
One owner
From the decision in the room to the system in production. The strategy and the operators are not two projects.
Evidence before scale
Traces, evaluations, and boundaries before the agent touches real work. A pilot that cannot be audited is not a result.
Scope from the risk, not the vendor menu
Boundaries, traces, and evaluations come from the risk of the workflow, before a vendor is chosen.
An owner, traces, evaluations, and boundaries. That is the standard before anything is called production.
Latest Writing
For the executive conversation
Supervisors Are Now Asking Who Owns the System
On 19 October 2026 the FCA published good and poor practice for AI in financial services. The failure it describes is familiar: governance fitted on after the design was already chosen. The same month, supervisors in Singapore and Hong Kong asked for a named owner.
Read essay →How I think about the work
Three questions a leadership team should be able to answer
This site is how I think about the work. It is not a PwC offering, and it is not a menu of private engagements.
- What the system is allowed to do
- What evidence exists of what it actually did
- Who owns it when behavior changes