KYC agent deployed on AWSHyderabad

On the record.

Governed AI that notices what doesn't match, reasons it through, and puts a recommendation on the record. Your team decides.

Cognance builds governed AI decision agents for India's NBFCs. Our KYC agent resolves the ambiguous name mismatches legacy vendors push into human queues, returning a recommendation with a confidence score and reasoning, typically in under a second, so your team reviews only the cases that genuinely need a human.

Run a shadow-mode pilot How it works

Reasoning receipt

REC-0042 · illustrative

Name on PAN: MOHD IRFAN SHAIKH
Name on Aadhaar: Mohammed Irfan Shaikh
Reasoning: "Mohd" is a standard abbreviation of "Mohammed". Given name and surname otherwise match exactly.
Confidence: 97.0

Logged with reasoning, version, timestampRECOMMEND APPROVE · YOUR DECISION

The decision gap

Legacy vendors extract data. They dump the ambiguity into your queue.

Name-match tools like the incumbents read a PAN and an Aadhaar name — then, the moment the two don't line up exactly, they eject the case to a human reviewer. That manual gap is the real cost: slow onboarding, expensive review teams, and compliance exposure.

Legacy KYC utilities

Extract fields, then eject anything ambiguous to a human review queue. You keep paying for the review team — and onboarding stalls while cases wait.

Every ambiguous case costs reviewer time and slows onboarding.

Cognance

Accepts the pre-extracted names and resolves the ambiguity using governed LLM reasoning. Returns a recommendation, with a confidence score and a reasoning log, via webhook.

Your team reviews exceptions, not every mismatch.

How it works

One API call in. A governed recommendation out.

A deterministic fast path resolves exact matches instantly. Only genuinely ambiguous cases reach the reasoner — and anything below the confidence threshold is routed to human review, never auto-rejected.

Client LOS

Authenticated call

Managed API

Auth + ingest

Durable queue

Reliability buffer

Decision engine

Governed reasoning

Amazon Bedrock

Model when needed

Audit store

Complete audit record

what you send

// One authenticated call from your LOS
{
  application_id: "LOAN-12345",
  pan_name: "Mohammed Khan",
  aadhaar_name: "Mohd. Khan"
}
 
// Two name strings. Nothing else.

webhook · decision callback

{
  "event": "kyc.decision",
  "application_id": "LOAN-12345",
  "status": "approved",
  "confidence_score": 95.0,
  "reasoning": "'Mohammed' and 'Mohd.'
     are common abbreviations…",
  "execution_time_ms": 455
}

Status is a recommendation. Your team retains the decision.

Read the technical deep-dive

Live today

Not a prototype. A deployed system.

The KYC agent runs today on AWS — a serverless pipeline behind an operational dashboard. These are measured facts from a small test set, not projections.

~8 ms

Fast-path decision

Exact match, no LLM call

~455 ms

LLM reasoning path

For ambiguous name pairs

Golden-dataset examples

Labeled examples today; benchmark pending a larger, balanced set

~$0.0003

Cost per verification

Calculated from AWS pricing

The data moat. With the client's consent, reviewed cases are masked and added to a labeled dataset that improves the agent over time. It holds 46 examples today, so accuracy benchmarking begins once it reaches a statistically meaningful, balanced size.

The operations console

Every decision, visible and auditable.

Operators review, override, and audit from a live dashboard — the same governance surface a compliance team uses at audit time.

console.cognance.in / operations

Processed

105

Auto-resolved

54.3%

Pending review

Golden set

Throughput · 7-day

Reasoner

Fast path~8 ms

LLM path~455 ms

status: approved
confidence: 95.0
path: fast_path

Illustrative representation of the operations dashboard. Figures are from an intentionally edge-case-heavy test set and are not production rates.

Who's building this

Two founders: a builder and a scaler.

Cognance is led by a two-person founding team. Our CEO built and deployed the KYC agent end to end; our CTO brings over a decade of data-engineering and delivery leadership to scale it into a governed decision platform for the NBFC sector.

2 founders

A builder and a scaler, ~23 years combined engineering and delivery experience

10+ yrs

10+ years in software and technology delivery, including financial-services modernization (CEO)

12+ yrs

Data engineering & delivery leadership (CTO)

Hyderabad

Where Cognance is based

Meet the founders

The roadmap

KYC is the wedge. The suite is the moat.

KYC establishes trust inside a client. From there, the same governed decision infrastructure extends to the other back-office judgment calls that still bottleneck regulated lenders.

Deployed

KYC Agent

Name-match recommendations. Deployed on AWS.

In development

Credit Underwriting

Credit memo recommendations for your underwriters.

Planned

Compliance Monitor

Continuous, automated compliance scanning.

Planned

Collections Intelligence

Recovery prioritization and outreach.

Security & compliance

Designed for audit.

Governance is the product. The compliance layer — isolation, audit, confidence routing, human fallback — is what separates a governed agent from a raw model call.

Data minimization

We receive only two name strings — no Aadhaar or PAN numbers, no DOB, no addresses, no documents, no biometrics.

7-day PII lifecycle

Name strings are scrubbed after 7 days. The non-PII decision record is retained for 5 years to satisfy PMLA.

Complete audit trail

Every decision logs confidence, reasoning, timestamp, and model version — queryable for audits.

Never auto-rejects

Uncertain cases route to human review. A dead-letter queue captures failed messages so nothing is silently dropped.

Regulatory mapping

Designed with reference to DPDP, the RBI KYC Master Direction and PMLA. SOC 2 is planned.

Data residency

Hosted on AWS Asia Pacific (Sydney) today. India-region (Mumbai) hosting is planned before any client processes production data.

See the full security & compliance overview

FAQ

Questions a compliance team asks first.

What data do you actually receive?

Only an application reference and two name strings — the name on the PAN and the name on the Aadhaar. No ID numbers, no dates of birth, no addresses, no documents, no biometrics.

What happens when the AI is unsure?

Cases below the confidence threshold are routed to human review, never auto-decided. During a pilot, the agent runs in shadow mode alongside your team, so your existing process remains the decision of record throughout.

How long does integration take?

About a day of client engineering: one authenticated API call and a webhook. No SDKs, no VPN, no database changes, no cloud lock-in.

Where is our data stored, and for how long?

On AWS Asia Pacific (Sydney) today; India-region (Mumbai) hosting is planned before production data. Name strings are scrubbed after 7 days; the non-PII decision record is retained for 5 years for PMLA.

How does pricing work?

A platform fee plus per-verification usage, with no setup fees. Pilots run free (30 days, up to 10,000 verifications). Commercial pricing is scoped per client after the pilot proves value.

Risk-free pilot

Run 10,000 verifications, free.

We integrate in about a day, run in shadow mode alongside your team, and only go live if accuracy meets an acceptance bar we agree with you up front. Best fit: NBFCs and fintech lenders processing 5,000+ KYC verifications a month on an LOS or core system with a REST API.

~1 day of client engineering
Shadow mode: parallel to your manual team
You keep your brand — we're invisible backend infrastructure