01 / Reject
Identify and reject the defaulters.
Find the borrowers your rules engine passes but who won’t repay—at the offer, before the money leaves, not in the NPA report.
Before disbursal, not at DPD 90
MentatAI-Native Underwriting Intelligence
By TARA AI LabsMore approvals.
Fewer defaults.
Every loan priced
to its risk.
Mentat, our AI underwriting model, reads the full repayment history behind a score and calculates default risk across every amount, rate and tenure at once. Approve the good borrowers your rules turn away, catch the risky ones they let through. Your policy still sets the ceiling.
Product lens
Prime · EMISynthetic demo
₹10k₹2L
Within PD ceiling
2.95%PD
Illustrative PD · 8% ceiling · Not a credit offer
Built on the depth of Indian credit.How these were measured
33Maccounts
Loan accounts in training
1.5Bevents
Repayment events in training
0.84AUC
Reported model discrimination
The film · 1:37
Money moves in real time.
Credit has to catch up.
Credit still waits for a risk decision while the payment clears. The film walks through the risk surface Mentat computes for every amount and rate, and why one model can price every product. Scored to Beethoven’s Egmont Overture.
01 / What Mentat does
Three levers.
One risk curve.
Reject the borrowers who won’t repay. Lend more to the ones who will. Price your best borrowers down—dynamically.
02 / Lend more
Raise disbursal inside your risk appetite.
Approve the safe borrowers a score cutoff turns away, and size every ticket up to your PD ceiling—not one flat amount for everyone.
Thin file is not the same as risky
03 / Price down
Dynamic rates for your best borrowers.
Rates follow each borrower’s PD. Your best borrowers get the rate they have earned, so they take the offer instead of dropping off to a competitor.
Good borrowers stop subsidising bad ones
02 / Rules vs observed risk
Rules run on assumed risk.
Mentat runs on observed risk.
A rules engine is a policy. Mentat is a model of what borrowers actually did. Your policy still sets the ceiling.
Business rules engineMentat
Where risk comes fromCutoffs a credit team wrote once—bureau ≥ 700, FOIR ≤ 50%Observed repayments—33M loan accounts and 1.5B repayment events in credit bureau data
What it readsOne bureau score and a checklistThe month-by-month repayment sequence behind the score, and cash flow
How the amount is setOne flat ticket, or a multiple of incomeA default probability for every amount—the offer is the largest one inside your ceiling
How it learnsWhen someone rewrites the rulesRetrained on fresh credit bureau data—never on a lender’s book
When you find outIn the NPA report, 90 days laterAt the offer—before the money leaves
03 / Make the decision visible
The whole curve.
Every product, together.
Keep all three products in view. Move the exposure, choose a borrower cluster or change the PD ceiling to see exactly where each curve crosses it.
◇ Curve intersections mark where the selected PD ceiling is reached.Full probability range. Curves remain visible across the full exposure range.
Prime · Product-conditioned Si and Sa positions feed these same curves. The 3D maps use the selected exposure; this chart shows the whole exposure range.
₹10,000₹2,00,000
How does a score cutoff differ?
For the Prime example, an illustrative bureau cutoff of 700 passes the profile at a score of 771. An income multiplier would suggest ₹1,56,000.
The complete risk curves show a different question: how does the estimated default probability change at every amount, for every product? A PD ceiling is one input to a lending decision, not an approval by itself.
Synthetic profiles and illustrative probabilities. The 3D maps and curves use the same function. Comparisons hold the default definition and horizon constant; credit lines assume fully drawn exposure. Not Mentat model outputs or lending decisions.
04 / The risk matrix
Every borrower has a risk curve.
Lend at the right point on it.
Five synthetic borrowers, one bullet loan from ₹10,000 to ₹2,00,000. Colour is default probability. The marker is the largest amount inside your 8% PD ceiling.
Swipe to see each verdict →
Default probability, low → high Above your PD ceiling Mentat offer Rules-engine amount
Rules engine · same five borrowers
3 of 5 approved · ₹3,75,000 out · 3 of 3 above the PD ceiling
Mentat · same five borrowers
5 of 5 approved · ₹4,05,000 out · every loan inside the PD ceiling
Synthetic profiles and illustrative probabilities—the same function as the curves above, and it follows the PD ceiling you set there. The rules engine shown is illustrative: bureau score ≥ 700, three times monthly income. Not Mentat model outputs or lending decisions.
05 / The geometry of credit
The bureau generalises.
Mentat contextualises.
Two continuous borrower axes. A third axis for product categories. Colour expresses PD; the boundary shows the selected ceiling.
Si01 / Behaviour
The intent to repay.
Read the sequence behind the score. Month-by-month repayment behaviour brings a borrower’s credit history into focus.
Bureau & repayment sequences
Sa02 / Capacity
The ability to repay.
Understand the room to borrow. Cash-flow stability, disposable income and existing obligations shape repayment capacity.
Cash flow & commitments
P03 / Context
The product that fits.
A bullet loan, an EMI and a credit line ask different things of a borrower. Evaluate risk within each product’s structure.
Product, amount & repayment structure
06 / AI-native underwriting
A rules engine is as smart on day 300 as on day one.
Mentat keeps learning.
The credit bureau records every repayment across India. Mentat retrains on that data—never on a lender’s book—and every release ships behind the same endpoint.
The model so far
FoundationMentat’s credit bureau model
33M loan accounts · 1.5B repayment events
Sep 2026Live inside lender underwriting
Scoring live applications through one API
NowNext version training on fresh bureau data
Same API for every lender already integrated
Your first monthLive inside your policy
Your ceilings, your products. Your data never trains the model
A rules engine
Learns when a credit team rewrites it. Each new lender starts from zero.
Mentat
Learns from credit bureau data, never from your book. Your data is used to score your applicants—never to train the model.
07 / Compliance at scale
At scale, RBI asks you to justify every rate.
Mentat is your risk-pricing derivation.
Risk-based, dynamic pricing across every borrower profile—compliant as you grow, because every rate traces back to a default probability.
What RBI expects
Interest rate model
Rates set through a board-adopted model of cost of funds, margin and risk premium—with the approach to gradations of risk, and why different borrowers pay different rates, disclosed in the application and the sanction letter.
October 2024
Four NBFCs stopped from sanctioning loans. RBI cited their pricing —lending rates and spreads over cost of funds seen as excessive.
Key Fact Statement
The all-in APR is on paper for every retail loan since 1 October 2024. Every borrower—and the regulator—sees the spread.
The derivation, borrower by borrower
Same cost of funds, operating cost and margin. The risk premium comes from each borrower’s PD—so the gap between two rates is explained, not asserted.
PrimePD 2.1% at ₹25,000
EmergingPD 4.1% at ₹25,000
New-to-creditPD 6.3% at ₹25,000
Cost of fundsOperating costRisk premium = PD × lossMargin
What your board, auditor and RBI can see
A PD for every borrower, behind the risk premium in their rate · the reasons behind every PD · a monotonic, explainable model · every model version tested before release
Sources: RBI Master Direction (NBFC – Scale Based Regulation), Fair Practices Code—interest rate model and gradations of risk · RBI press release on action against four NBFCs, 17 October 2024 · RBI Key Fact Statement circular, April 2024 (effective 1 October 2024). Bars show structure only, using the synthetic borrowers above.
08 / Built for your lending stack
Intelligence in.
A decision you can use.
Bring bureau and cash-flow inputs together. Return product-aware risk and offer parameters your team can evaluate against its own policy.
Product-specific assessments
Risk and capacity in one response
Your policy remains the reference
assessment.json
EXAMPLE product-aware assessment
{
"borrower": "SYNTHETIC_1",
"product": "emi",
"assessment": {
"requested_amount": 100000,
"illustrative_pd": 0.0295,
"policy_ceiling": 0.08,
"within_policy": true
},
"sizing": {
"illustrative_limit": 200000,
"currency": "INR"
},
"simulation": true
}09 / The global direction
Global lenders already underwrite on behaviour.
Short-term loans gain the most.
Block reports that its Cash App Score approves more borrowers at the same risk. Mentat brings the same idea to Indian credit.
United States · Block · Cash App Score
+38%
more customers approved at the same level of risk, on short-term Borrow loans.
+30%auto loans approved at identical loss rates
+28%credit card applications approved at identical loss rates
India · TARA AI Labs · Mentat
Same idea.
Built for the decisions Indian lenders make.
- Learns from observed repayment behaviour, not fixed rules
- 33M loan accounts and 1.5B repayment events in training
- Bullet loans, EMI and credit lines underwritten as separate products
- Runs on the bureau and cash-flow inputs you already pull
Block figures quoted from block.xyz/cash-app-score; Block notes they reflect historical performance. Block is not affiliated with TARA AI Labs.
10 / Research behind the engine
Depth in the data.
Clarity in the evidence.
Mentat learns from credit bureau data, never from a lender’s book. Start with the reported results, then evaluate performance on the segment and outcomes that matter to your book.
Request the evaluation methodologyReported AUC0.84
Default concentration in riskiest decile8×
Training loan accounts33 million
Training repayment events1.5 billion
Internal evaluation by TARA AI Labs on held-out Indian retail credit data. Not independently audited, and not a result from any lender’s book. The evaluation population, period, outcome definition and baseline are set out in the methodology note — ask for it, and read it, before you rely on these figures.
Your portfolio. A clearer picture.
Find the opportunity
inside your book.
Start with one portfolio segment. Explore where risk, loan sizing and your current policy tell different stories.
Share your lending context. The Mentat team will follow up with the process and data requirements.
Or book a walkthroughOn the horizonFacial KYC. Identity verification before underwriting.Ask about the roadmap