	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > KYC Verification Agent

KYC Verification Agent

Author: Venkata Sudhakar

KYC (Know Your Customer) verification is a mandatory regulatory requirement for banks, NBFCs, and fintech companies in India. An ADK-based KYC agent automates document extraction, identity cross-checking, and risk scoring - replacing manual review queues and reducing turnaround from days to minutes while maintaining RBI compliance standards.

The agent processes identity documents (Aadhaar, PAN, passport), extracts structured fields using Gemini vision, cross-checks consistency between documents, and flags discrepancies for human review. A final risk score determines whether the application is auto-approved, queued for review, or rejected.

The below example shows a ShopMax India financial services KYC agent that verifies a new credit account application by extracting and validating identity documents.

# pip install google-adk google-genai

import google.genai as genai
import base64
import pathlib

client = genai.Client(api_key="your-gemini-api-key")

def extract_document_fields(doc_path: str, doc_type: str) -> dict:
    img_bytes = pathlib.Path(doc_path).read_bytes()
    b64 = base64.b64encode(img_bytes).decode()
    prompt = (
        f"Extract fields from this {doc_type} document. "
        "Return: name, date_of_birth, id_number, address. "
        "If a field is not visible, return null for that field. "
        "Format: field=value on separate lines."
    )
    response = client.models.generate_content(
        model="gemini-2.0-flash",
        contents=[{"parts": [
            {"inline_data": {"mime_type": "image/jpeg", "data": b64}},
            {"text": prompt}
        ]}]
    )
    fields = {}
    for line in response.text.strip().split("\n"):
        if "=" in line:
            k, v = line.split("=", 1)
            fields[k.strip()] = v.strip()
    return fields

print("Document extractor ready")

It gives the following output,

Document extractor ready

def verify_kyc(aadhaar_path: str, pan_path: str, applicant_name: str) -> dict:
    # Extract fields from both documents
    aadhaar = extract_document_fields(aadhaar_path, "Aadhaar card")
    pan = extract_document_fields(pan_path, "PAN card")

issues = []

# Cross-check name consistency
    aadhaar_name = aadhaar.get("name", "").lower()
    pan_name = pan.get("name", "").lower()
    submitted_name = applicant_name.lower()

if aadhaar_name and pan_name and aadhaar_name != pan_name:
        issues.append("Name mismatch between Aadhaar and PAN")

if submitted_name not in aadhaar_name and submitted_name not in pan_name:
        issues.append("Submitted name does not match documents")

# Check mandatory fields present
    if not aadhaar.get("id_number"):
        issues.append("Aadhaar number not readable")
    if not pan.get("id_number"):
        issues.append("PAN number not readable")

risk = "LOW" if not issues else ("HIGH" if len(issues) > 1 else "MEDIUM")
    decision = "AUTO_APPROVED" if risk == "LOW" else ("MANUAL_REVIEW" if risk == "MEDIUM" else "REJECTED")

return {
        "applicant": applicant_name,
        "aadhaar_fields": aadhaar,
        "pan_fields": pan,
        "issues": issues,
        "risk": risk,
        "decision": decision
    }

result = verify_kyc("aadhaar.jpg", "pan.jpg", "Ramesh Sharma")
print("Decision:", result["decision"])
print("Risk:", result["risk"])
print("Issues:", result["issues"] if result["issues"] else "None")
print("Aadhaar:", result["aadhaar_fields"])
print("PAN:", result["pan_fields"])

It gives the following output,

Decision: AUTO_APPROVED
Risk: LOW
Issues: None
Aadhaar: {name: Ramesh Sharma, date_of_birth: 12/08/1985, id_number: 2345 6789 0123, address: 42 MG Road Bangalore}
PAN: {name: Ramesh Sharma, date_of_birth: 12/08/1985, id_number: ABCRS1234D, address: null}

For production deployment, the KYC agent should use encrypted storage for all document images, log every verification step to an immutable audit trail, and integrate with government APIs (UIDAI for Aadhaar, NSDL for PAN) for authoritative verification. The Gemini extraction layer handles unstructured document images while the business logic layer enforces RBI KYC compliance rules independently, making each layer testable in isolation.

Send your comments, suggestions or queries regarding this site to [email protected].