PhonePe control brief Risk angle · AI at scale

Their world

PhonePe already runs AI in production — Agent Hub watches its own dashboards, and merchant onboarding now happens through a conversation.

The next move is not adding AI. It is holding the line on control as AI spreads through a regulated platform at scale, with an approved IPO now in front of you.

Their pressure

Every AI-touched workflow is now a workflow the market will audit.

At tens of millions of daily transactions across 560 million users, the distance between a helpful model and an unreviewable one is a single missing control.

The bridge

Meridian AI puts the controls on before AI reaches the transaction path.

We are a boutique applied-AI studio that ships production-grade LLM and agent systems on your data, with human-in-the-loop controls, in six weeks — not six quarters. For PhonePe that means every AI action can start from an approved input, produce a record your systems and reviewers can inspect, and stop for a person before it touches a regulated flow.

Audit trail · one AI-touched action

  • Request scopedApproved input and permitted task defined up front.
    Bounded
  • Decision recordedStructured output your systems and reviewers can inspect.
    Logged
  • Guardrail enforcedOut-of-policy actions blocked; evals run in CI before deploy.
    Checked
  • Human sign-offExceptions stop for a person before a regulated flow.
    Held

One proof

We ship on your existing stack — retrieval over your own data, function-calling agents wired into your current services, and evals plus guardrails in CI. No model training, no rip-and-replace, no new infrastructure to run.

One working session

Trace one PhonePe AI workflow from input to audit record.

Twenty minutes on one live workflow — merchant onboarding, search, or ops — mapping the task boundary, the guardrail, and the human-review path your reviewers can sign off.

See the 20-minute teardown for PhonePe