Data pipelines, models, strategies, and provisions — the risk team's entire workshop on one governed fabric, with an AI copilot that answers everyone on it.
In commercial engagement with South Indian Bank and Jana Small Finance Bank.
Real-time verdicts are served by NirnayaDecisionAI — same models, same lineage, paired deployment.
Cataloguing, modelling, ETL, BI, and decisioning from five vendors means every policy change is a project. On one fabric, it's an afternoon.
One ontology, end to end. Because Pragna shares the same 1,500-entity lending ontology as AarambhLOS and Nirnaya, the borrower your origination system knows is the same borrower your early-warning model tracks. No ETL. No reconciliation. No Sunday-night spreadsheet before the board meeting.
One policy of record. Model governance, validation packs, and immutable audit — examination-ready by construction, not by scramble.
ECL staged on every account, scenario-weighted, reconciled to the same data the business runs on. Board pack out of the system, not out of Excel.
Build the tree, see P&L per node, test on live traffic, promote in a click. Weeks of change requests become an afternoon of work.
Ask Buddhi answers portfolio questions in seconds — grounded in governed definitions, so every team quotes the same number.
Adjusting a housing-loan policy typically takes 6–10 weeks across IT, risk, and BIU. In Pragna, strategy teams build the tree, optimise the cut-offs, simulate the P&L, and promote it themselves — with economics attached to every node.
AutoGrow simulation re-evaluates the full tree on a 6-month hold-out before anything ships. Illustrative figures.
Build the tree visually — or describe the policy and let the copilot draft it with economics per node.
Replay 6 months of bookings; see per-node approval, expected loss, NIM, and RAROC deltas.
Champion–challenger on 5–10% live traffic; 1-click promote with audit trail, auto-rollback on breach.
Staging, PD × LGD × EAD, and macro-weighted scenarios run wired into the same fabric as origination and monitoring — so provisions move when the book moves, not when the quarter ends.
Why this matters now: April 2027 Is Not an Accounting Deadline →
Illustrative quarter-end run.
Grounded in a 1,500-entity ontology · governed SQL · every Q&A logged for audit
Portfolio questions that took the MIS team two days come back in seconds — because Buddhi doesn't guess. Natural language becomes governed SQL through an 11-product, 1,500-entity semantic ontology, so every team computes the same number from the same definitions.
Natural-language Q&A scoped to governed Spaces per team.
Auto-charts: vintages, roll rates, drift — rendered on the spot.
Drafts board narratives and MRM packs from live data.
Analysts verify queries and publish them as one-click prompts.
In 8–12 weeks we can scope, build, and shadow-run a housing-loan acquisition strategy on your data — with full ECL roll-up.
