The Underwriting Engine
Best-in-class credit risk.
Live today.
Sentinel treats underwriting as geometry — every borrower a vector, every product its own risk surface. Built on BASIC V4, the transformer that reads the repayment sequences every other model throws away.
0.84 AUC
~45% error reduction over aggregate-feature baselines
8x
default concentration caught in the riskiest decile
33M
loan accounts · 1.5B repayment events in training
2,000+
cashflow variables read per income statement
Three Axes
One vector space. Three dimensions.
Intent to Pay
X-Axis · Bureau Prediction
BASIC V4 reads month-by-month repayment sequences other models discard — trained on 33M accounts and 1.5B repayment events.
Ability to Pay
Y-Axis · Cashflow Underwriting
2,000+ variables from bank statements, salary slips, UPI and GST — fraud, stability, disposable income. A cashflow signature, not a number.
Product Construct
Z-Axis · The Core Axis
Bullet loan, EMI and credit line have different physics. The same borrower gets a different vector, offer and risk score per construct.
Built for NBFCs
What Sentinel does to your book.
01
Cut NPAs before disbursal
With 8x default concentration, the riskiest decile is visible on day zero — price it, cap it, or decline it before the loan exists.
02
Collapse underwriting ops cost
Pre-approved offers replace manual queues. One API call, decision in ~1.2 seconds, zero-touch for the vast majority of files.
03
Native to your board policy
Offers are generated inside your approved risk policy — monotonic constraints, RBI-explainable, audit-ready scores at 30/60/90 DPD.
04
A portfolio that compounds
With every repayment, the real-time feedback loop re-tunes underwriting to your book — approvals sharper, pricing tighter, portfolio quality compounding month after month.
The Model Suite
Underwrite. Retarget. Recover.
LIVE
Core Underwriting
Per-construct risk vectors, pre-approved offers and PD scores for every verified KYC user — the engine running on this page.
INDUSTRY FIRST
User Credit Affinity
Predicts which user will need credit, when, and for which construct — bullet, EMI or credit line — so you retarget with the right offer before they ever go shopping. Nobody else models this.
IN PILOT
Sentinel Triage
Continuously re-scores your live book against your default-risk policy and ranks where recovery effort goes first — focused intervention on accounts drifting toward default, light touch everywhere else.
The most accurate retail credit model in India today.
BASIC V4 outperforms aggregate-feature scorecards by ~45% on ranking error. If your current bureau strategy is a scorecard, Sentinel sees risk it structurally cannot.
Per Verified KYC User
What the API returns.
Pre-Approved Offer
Optimal product, amount, tenure, rate — mapped to the lender's risk appetite. A deployable offer, not a score.
Default Risk Score
PD at 30/60/90 DPD. Monotonic constraints. RBI-explainable and auditable.
Cashflow Capacity
Disposable income, stability index, fraud flags. The Sa vector for policy integration.
GraphQL API Response
// per-product vector output
{
"user": "KYC_7291",
"product": "emi_personal",
"vector": { "si": 0.87, "sa": 0.72 },
"offer": {
"amount": 150000,
"tenure": 12,
"rate": 16.5
},
"risk": {
"pd_30": 0.031,
"pd_90": 0.008
}
}
The Core Goal
Better collections.
Lower NPAs.
Lower NPAs.
Right product, right amount, right risk tier — for every verified KYC user. Lower defaults, lower ops cost, and a real-time feedback loop that never stops learning from outcomes.
Start with a free portfolio analysis.
The Sentinel team runs your book — one segment is enough — and shows you the risk you're not pricing, the users worth retargeting, and where recovery effort actually moves the needle. Free, and the findings are yours to keep.