Skip to content
Reference vertical · Banking & NBFCs

Agentic AI for banks, NBFCs, SFBs

Loan underwriting. AML alert review. KYC enhancement. Exception triage. Four workflows where banks burn the most reviewer hours per crore of business — and where regulator audit expectations make automation feel risky. The Vihaya Engine solves the audit primitive first, then automates. India-context regulatory framing covered. Pre-revenue; no paid pilot has completed.

Designed for RBI
Master Direction on IT Outsourcing 2023
Reference
Solution architecture in repo · pilots open
Designed for VPC
AWS Mumbai · GCP Mumbai · Azure South India

Workflows in scope

Retail loan underwriting

Read the application, the bureau report, the income proofs, and the policy. Decide approve / decline / refer with citations. Personal, two-wheeler, gold, agri-credit, education.

AML alert review

Read the alert, the transaction history, the customer profile, the FIU-IND typology guidance. Output: clear / refer / SAR-recommend with cited rationale.

KYC re-verification

Compare KYC documents against the existing record. Flag re-verification needs grounded in the bank’s CDD policy + RBI KYC Master Direction.

Trade-finance exception triage

Cross-document reconciliation (LC, invoice, BL, packing list) for trade-finance exception handling with cited discrepancies.

Why RBI alignment is the unlock

Bank technology teams have been able to automate decisioning for years. The blocker has been the audit chain. RBI examiners want to walk into a bank, pick a specific loan-decline from three years ago, and see exactly which policy clause, which bureau report line, and which exception path produced that decision. Most ML systems can’t answer that question.

The Vihaya Engine’s audit primitive is designed for exactly that: every decision is one immutable record linking to the policy clauses cited, the bureau-data points referenced, the agent run that produced it, and the human reviewer who confirmed or overrode it. Reconstructable from cold storage years later. That’s the ask that turns ’AI for banks’ from a vendor pitch into a deployable system.

Products in scope (illustrative)

Pilot scope per product. Touchless-rate targets are tuned during each pilot’s shadow run; Vihaya is pre-revenue and no paid pilot has yet completed, so we don’t publish ranges as fact.

ProductPilot scopeTouchless rate
Personal loanSalaried, ticket-size ≤ ₹5LTuned per pilot
Two-wheeler loanSalaried + self-employedTuned per pilot
Gold loanFirst-cycle and renewalTuned per pilot
AML L1 reviewSanctions / threshold / typology alertsTuned per pilot
KYC re-verificationPeriodic CDD refreshTuned per pilot

Banking & NBFC FAQ

Does Vihaya satisfy RBI’s IT outsourcing framework?

Vihaya is designed against RBI’s 2023 Master Direction on Outsourcing of IT Services, which requires data localisation, examiner audit rights, exit-management plans, and BCP/DR provisions. The intended deployment is inside your bank’s VPC (AWS Mumbai, GCP Mumbai, or Azure South India) — no data egress to Vihaya. The audit trail is built to support examiner inspection of any specific decision. The draft SOW we’d propose includes the right-to-audit clause and exit-management plan. Final compliance posture is the bank’s determination, not Vihaya’s certification.

What about cooperative banks and small-finance banks?

Same engagement shape. SFBs and large urban cooperative banks have the volume to justify the pilot economics — we scope pilots for banks running large retail-credit volumes per month. The pilot is structured around one product (personal loan, two-wheeler loan, gold loan, agri-loan) before extending.

Can Vihaya read CIBIL, Experian, Equifax reports?

The architecture ingests structured credit-bureau reports alongside the unstructured policy and application text. The agent’s decision rationale would cite both the bureau-data point and the policy clause it triggered.

How does this compare to existing decisioning engines like FICO Origination Manager?

Rule engines like FICO OM are still excellent for hard-policy filters (income thresholds, score cutoffs). Vihaya is designed to layer on top — handling the document-parsing, judgement-grade evaluations, and exception cases that rule engines cannot decide. We expect rule engines to keep handling deterministic policy filters while Vihaya layers on the judgement-heavy and document-parsing work; the split is tuned during the pilot.

Is this safe for AML alert review under PMLA?

AML review under the Prevention of Money Laundering Act and FIU-IND guidance requires explainable, audited decisions. Vihaya’s primitive is designed to match: every alert disposition is one immutable row with citations to the underlying transaction patterns and policy bulletins. Below the confidence floor, the case routes to your compliance analyst — never auto-cleared.

Next step

Want to see this in your environment?

30-minute discovery call. We follow up with a draft SOW shortly after.

Talk to us about a pilot