Onboarding and origination, run for you · Digital lending and NBFCs

We run onboarding and loan origination. Compliant approval, first session.

Antino builds an onboarding brain — your KYC, credit, and compliance rules as an executable graph — and runs regulated onboarding and origination on it. Every verified user and every originated loan carries its own compliant audit trail.

Outcome: compliant approval in the applicant's first session.Proof: Revfin

Below: the whole pipeline — playable

01 · The problem

Compliance and conversion are pulling in opposite directions.

Every regulatory tightening adds a step, and every step drops applicants at the door. eKYC, vKYC, and co-lending logic live across vendors and policy memos, so a rule change means a sprint and a prayer. You either over-collect and lose users or under-check and inherit risk. Both are expensive.

02 · How we'd run it

The onboarding brain: every applicant in, two ways out.

Intelligent decisions. Zero busywork. Every exception, owned.

Applicants in

Night and day. Mostly night.

Identity documents
PAN, Aadhaar, DigiLocker
Live selfie & device
Face liveness, device signals
Credit & bureau data
Score, history, obligations
Lender policies
Each co-lending partner's rules

The onboarding brain

Verify. Decide. Disburse.

Document verification
Face liveness
KYC rules
Credit policy
Co-lending routing
Sanctions & blacklist

every decision logged · every outcome written back

Straight-through

~90%

Verified and approved while the applicant is still in the flow.

MinutesFully compliantNo office hours

Exception

~10%

Routed to a vkyc reviewer

Complex cases. Human judgment — with the case pre-packaged:

Full document setLiveness clipRule triggered

Executable KYC rules

Regulation as decision tables

Auditable by design

Every verification logged

Continuous learning

Each review tightens the rules

Regulator ready

RBI-grade trails & checkpoints

Scales by default

Volume spikes without queues

Shares are illustrative of a typical mature origination flow.

03 · The pipeline, end to end

The science of shipping, stage by stage.

Nine stages from a 10:40 pm tap on an ad to a disbursed, regulator-ready loan. Drag the agentic-depth dial to see how much of the regulated path runs itself at each depth, and hover any stage for a worked example. Two things never automate: the credit override and the final say on a borderline identity.

Agentic depth — drag it

≈ 0% of routine work absorbed1 human-owned gates — at any settingtick on each drop = that stage's honest ceiling; past it stays human

work units on the trackhuman gate — the flow always passes throughabsorbed into the agent lane

Verify

Onboarding brain · verify0%

stage 01 · dossier

Application intake

agent

The application starts on a phone; every consent is captured inline at the moment it's legally required.

Orchestrates the regulated journey as one flow: PAN, Aadhaar, documents
Captures each consent in-flow, stamped, never reconciled later

An EV-loan applicant, from an ad, at night

A driver taps through at 22:40. The flow asks for exactly what regulation requires, exactly when it requires it — and the consent trail writes itself as he goes.

A form that assumes office hoursA regulated front door, open at 10:40 pm

click the stage to collapse

stage 02 · dossier

Document verification

agent

PAN, Aadhaar and DigiLocker documents are verified at source and cross-checked against each other.

Verifies documents against issuing sources, not against photocopies
Cross-checks name, DOB and address across every document submitted

The mismatch that surfaces in seconds

A PAN-Aadhaar name variant is detected and normalized in-flow. The applicant never learns there was a check; the audit trail records that there was.

A back-office checks photocopiesCross-checked at source in seconds

stage 03 · dossier

Face liveness

agent

Liveness detection confirms a live human matches the verified document — at night, on a phone, without an office.

Runs face-liveness and document-face match on the live selfie
Writes the liveness evidence into the audit trail as it happens

The pattern from regulated EV lending

Face-liveness built for real onboarding conditions — low light, cheap phone cameras, applicants in a hurry — so the clean applicant passes in seconds and the spoof attempt doesn't.

A branch visit to prove you existA selfie, scored for liveness, in-flow

stage 04 · dossier

Dedupe, AML & sanctions

agent

Dedupe, AML and sanctions checks run inline; each one writes its evidence as it completes.

Runs dedupe, blacklist and sanctions screens in-flow
Logs every screen's evidence for regulator replay

Verified in two minutes, replayable forever

Every check that fired, what it consulted and what it concluded — written at execution time, so a regulator can replay the whole verification step by step.

Screens run in overnight batchesScreened before the next tap

stage 05 · dossier

vKYC review

agenthuman

Borderline matches go to a human reviewer with the gap pre-flagged — never silently passed, never restarted.

Routes the case with the exact gap highlighted: the score, the clip, the document set
A vKYC reviewer clears or rejects the match. The threshold never loosens itself.

Name-match 0.71 — close, not clean

The applicant isn't bounced back to the start. The morning reviewer sees the flagged gap, the liveness clip and the documents side by side, and clears it in one look.

Re-do the journey, in office hoursOne look at the flagged gap

stage 01 · dossier
Application intake
agent
The application starts on a phone; every consent is captured inline at the moment it's legally required.
Orchestrates the regulated journey as one flow: PAN, Aadhaar, documents
Captures each consent in-flow, stamped, never reconciled later
An EV-loan applicant, from an ad, at night
A driver taps through at 22:40. The flow asks for exactly what regulation requires, exactly when it requires it — and the consent trail writes itself as he goes.
A form that assumes office hoursA regulated front door, open at 10:40 pm
click the stage to collapse
stage 02 · dossier
Document verification
agent
PAN, Aadhaar and DigiLocker documents are verified at source and cross-checked against each other.
Verifies documents against issuing sources, not against photocopies
Cross-checks name, DOB and address across every document submitted
The mismatch that surfaces in seconds
A PAN-Aadhaar name variant is detected and normalized in-flow. The applicant never learns there was a check; the audit trail records that there was.
A back-office checks photocopiesCross-checked at source in seconds
stage 03 · dossier
Face liveness
agent
Liveness detection confirms a live human matches the verified document — at night, on a phone, without an office.
Runs face-liveness and document-face match on the live selfie
Writes the liveness evidence into the audit trail as it happens
The pattern from regulated EV lending
Face-liveness built for real onboarding conditions — low light, cheap phone cameras, applicants in a hurry — so the clean applicant passes in seconds and the spoof attempt doesn't.
A branch visit to prove you existA selfie, scored for liveness, in-flow
stage 04 · dossier
Dedupe, AML & sanctions
agent
Dedupe, AML and sanctions checks run inline; each one writes its evidence as it completes.
Runs dedupe, blacklist and sanctions screens in-flow
Logs every screen's evidence for regulator replay
Verified in two minutes, replayable forever
Every check that fired, what it consulted and what it concluded — written at execution time, so a regulator can replay the whole verification step by step.
Screens run in overnight batchesScreened before the next tap
stage 05 · dossier
vKYC review
agenthuman
Borderline matches go to a human reviewer with the gap pre-flagged — never silently passed, never restarted.
Routes the case with the exact gap highlighted: the score, the clip, the document set
A vKYC reviewer clears or rejects the match. The threshold never loosens itself.
Name-match 0.71 — close, not clean
The applicant isn't bounced back to the start. The morning reviewer sees the flagged gap, the liveness clip and the documents side by side, and clears it in one look.
Re-do the journey, in office hoursOne look at the flagged gap

Decide & disburse

Onboarding brain · decide & learn0%

stage 06 · dossier

Credit decision

agent

Eligibility, bureau data, segment rules and limits execute as decision tables on the verified profile.

Executes credit policy as decision tables, policy version stamped on the offer
Pulls bureau and obligation data and computes the limit

The offer that knows which policy made it

Eligibility, segment rules and limit calculation fire on the verified profile. The generated offer carries its policy version — so when policy changes, history stays explainable.

Policy as the adjudicator remembers itPolicy as written, version-stamped

stage 07 · dossier

Credit override

agenthuman owns

Borderline scores and policy exceptions go to a named credit officer — the override is human, recorded, and rare.

Prepares the exception file: the rule that fired, the bureau picture, comparable past overrides
The credit officer owns every override. This gate is permanent.

An override with a memory

The officer sees the triggering rule and how similar overrides performed. Whatever she decides is recorded against the policy version — so exceptions inform the next policy, instead of eroding this one.

Exceptions argued over emailA built file on a credit officer's desk

stage 08 · dossier

Co-lending allocation

agent

The co-lending allocation resolves automatically and partner records sync at the moment of origination.

Resolves the partner split per the encoded arrangement
Syncs partner records at origination, not at month-end

The 80/20 that nobody reconciles

The pattern from running co-lending rails: the allocation resolves as the loan books, both partners' records agree by construction, and month-end reconciliation has nothing to find.

Partner splits reconciled monthlyThe split resolves at origination

stage 09 · dossier

Disbursal & funnel write-back

agenthuman

The loan lands with its full trail; drop-off and fraud patterns write back so conversion rises at constant controls.

Disburses with the complete trail: identity evidence, consents, policy version, decision path
Feeds drop-off and fraud patterns back into the flow design
Control changes are proposed by the brain, approved by compliance.

Conversion up, controls untouched

Funnel data shows applicants abandoning at the selfie step; the flow is redesigned and re-tested. More applicants get through the same checks — growth without a weaker gate.

Disbursed, then forgottenEvery journey teaches the funnel

stage 06 · dossier
Credit decision
agent
Eligibility, bureau data, segment rules and limits execute as decision tables on the verified profile.
Executes credit policy as decision tables, policy version stamped on the offer
Pulls bureau and obligation data and computes the limit
The offer that knows which policy made it
Eligibility, segment rules and limit calculation fire on the verified profile. The generated offer carries its policy version — so when policy changes, history stays explainable.
Policy as the adjudicator remembers itPolicy as written, version-stamped
stage 07 · dossier
Credit override
agenthuman owns
Borderline scores and policy exceptions go to a named credit officer — the override is human, recorded, and rare.
Prepares the exception file: the rule that fired, the bureau picture, comparable past overrides
The credit officer owns every override. This gate is permanent.
An override with a memory
The officer sees the triggering rule and how similar overrides performed. Whatever she decides is recorded against the policy version — so exceptions inform the next policy, instead of eroding this one.
Exceptions argued over emailA built file on a credit officer's desk
stage 08 · dossier
Co-lending allocation
agent
The co-lending allocation resolves automatically and partner records sync at the moment of origination.
Resolves the partner split per the encoded arrangement
Syncs partner records at origination, not at month-end
The 80/20 that nobody reconciles
The pattern from running co-lending rails: the allocation resolves as the loan books, both partners' records agree by construction, and month-end reconciliation has nothing to find.
Partner splits reconciled monthlyThe split resolves at origination
stage 09 · dossier
Disbursal & funnel write-back
agenthuman
The loan lands with its full trail; drop-off and fraud patterns write back so conversion rises at constant controls.
Disburses with the complete trail: identity evidence, consents, policy version, decision path
Feeds drop-off and fraud patterns back into the flow design
Control changes are proposed by the brain, approved by compliance.
Conversion up, controls untouched
Funnel data shows applicants abandoning at the selfie step; the flow is redesigned and re-tested. More applicants get through the same checks — growth without a weaker gate.
Disbursed, then forgottenEvery journey teaches the funnel

Depth bars are illustrative of a mature origination flow. The credit override and the borderline-identity call are permanent human gates — the brain builds the file; it never owns the exception.

04 · One brain, two engines

How the function runs on top of the brain.

Revenue OS

Onboard and verify

We run regulated onboarding on the onboarding brain. eKYC and vKYC execute as a governed flow, consent and checks are captured inline, and a clean applicant is verified end to end without a manual handoff.

eKYC and vKYC orchestrated as one compliant, auditable journey
Identity, document, and liveness checks resolved straight through where clean
Consent and regulatory steps captured in-flow, not reconciled later
Suspect or incomplete cases routed to a reviewer with the gap flagged

Delivery OS

Originate and disburse

We run loan origination on the live brain. Credit policy executes against the verified profile, co-lending splits resolve automatically, and the originated loan lands with a full compliance trail.

Credit and eligibility policy applied as executable rules, not adjudicator habit
Co-lending logic and partner allocation handled inside origination
Each originated loan carries a complete, regulator-ready audit trail
Policy exceptions escalated to a human owner with the rule that triggered them

05 · What actually changes