Nirnaya means judgment. Rules, scorecards, ML models, and pricing — orchestrated into one defensible verdict, approve, refer, or decline, with the reasons attached and the audit record already written. Speed comes standard.
Live decisioning at South Indian Bank and Jana Small Finance Bank.
Illustrative console · every verdict ships with its reasons and an append-only audit record.
Illustrative outcomes, six months post-deployment on a housing-loan portfolio.
No translation layer. Every Nirnaya decision feeds directly into Pragna's portfolio monitoring and PramanaRiskAI's model governance — same borrower entity, same feature definitions, no reconciliation. The population you decide on is the population you provision for.
Why does explainability matter when the RBI examiner walks in? → · Read the production case study →
Every verdict carries reason codes and an append-only audit record. Policy is versioned, simulated before promotion, and reproducible bit-for-bit for any past date — examination-ready by construction.
Pricing and limits set per strategy-tree leaf, not per product. Change a cut-off, reprice a tier, or launch a segment offer the same day — without an IT change request.
Straight-through processing for clear cases; only genuine edge cases route to underwriters, with the full data picture attached. Fewer queues, faster files, cleaner hand-offs.
Machine intelligence runs through every layer of the decision — from the features it sees to the policies it executes.
Visual pipelines engineer hundreds of origination and behavioural features from LOS, core banking, and bureau data — 32M rows in minutes.
CIBIL, Experian, CRIF, and Equifax responses mapped into one canonical schema — tradelines, enquiries, and scores speak one language before models see them.
Interpretable WoE scorecards composed with gradient-boosted models — composite scores with SHAP-based reason codes on every decision.
Describe a credit policy in plain language; get a draft strategy tree with treatments and economics per node — ready to simulate, never auto-shipped.
New product or low-data portfolio? Simulate loss distributions across thousands of scenarios to stress cut-offs before a single loan is booked.
Route 5–10% of live traffic to a challenger, replay six months of bookings, compare P&L — promote in one click or auto-rollback on breach.
Risk teams describe what they want. The copilot drafts the strategy tree, attaches projected economics, and hands it back for review and simulation — humans approve, always.
Every data source, every rule, and every model converges into one under-300 ms, fully audited credit decision.
Approve / decline in under 300 ms
A new strategy never bets the book. Route 5–10% of live traffic to the challenger, replay six months of bookings against it, and compare the P&L — per node.
Illustrative shadow run · HL acquisition strategy.
See NirnayaDecisionAI decide a live application — bureau to verdict — in a 30-minute walkthrough.
