From insight to impact: See → Predict → Reason → Simulate → Decide → Act. The three strips of the architecture — the decision stack, the use cases, the agentic workforce — rendered bullet by bullet, each opening the thing that implements it, each badged with what it actually is.
27 named capabilities across six layers. 13 computed · 7 live AI · 3 views · 3 partial · 1 planned.
8 solution families, 29 capabilities. 14 computed · 9 live AI · 5 views · 1 partial.
Each agent carries the three skills the architecture names for it. Every skill is a distinct, runnable task over its own data slice, and every run writes to the decision log.
A diagram is cheap. The status on every bullet above is the point of this page.
Deterministic code over live records — scorers, rules engines, the simulator. Same input, same output, every factor shown.
Model reasoning grounded on computed evidence or governed data. The model explains or narrates; it does not make the decision.
A real page over real records, with no scoring or reasoning applied to it.
Exists, but narrower than the bullet claims. The note on the bullet says how.
Not built. Nothing on the page implies otherwise.
Two caveats that stay true across every page: the predictive scorers run on synthetic demo rows, so they demonstrate the mechanism rather than predictive power on a real book; and the combined ratio at 112% is an underwriting loss, covered by ₹598 Cr of investment income — never presented as profit.