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

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > Loan Underwriting Agent

Loan Underwriting Agent

Author: Venkata Sudhakar

Loan underwriting requires evaluating an applicant's income, credit history, existing liabilities, and repayment capacity against a lender's risk policy. A Gemini-powered agent can automate this evaluation - parsing application data, applying underwriting rules, and producing a structured approval or rejection with reasoning.

ShopMax India offers consumer finance for electronics purchases. Customers apply for EMI loans at point of sale. The underwriting agent evaluates each application in real time, reducing approval time from hours to seconds for clear-cut cases.

The below example shows a loan underwriting agent that assesses a personal loan application and produces a risk-graded recommendation.

import google.generativeai as genai
import json

genai.configure(api_key="your-api-key")
model = genai.GenerativeModel("gemini-2.0-flash")

# Lending policy rules
POLICY = {
    "min_credit_score": 650,
    "max_dti_ratio": 0.45,  # debt-to-income ratio
    "min_monthly_income": 25000,  # Rs
    "max_loan_to_income": 10,  # max loan = 10x monthly income
    "employment_types_allowed": ["salaried", "self_employed_stable"]
}

def underwrite_loan(applicant):
    # Pre-screen: hard eligibility checks
    if applicant["credit_score"] < POLICY["min_credit_score"]:
        return {"decision": "rejected", "reason": "Credit score below minimum threshold"}
    if applicant["employment_type"] not in POLICY["employment_types_allowed"]:
        return {"decision": "rejected", "reason": "Employment type not eligible"}

# Calculate DTI
    monthly_emi = applicant["requested_amount"] / applicant["tenure_months"]
    dti = (applicant["existing_emi"] + monthly_emi) / applicant["monthly_income"]

# Use Gemini for holistic risk assessment
    prompt = f"""You are a loan underwriting specialist for ShopMax India consumer finance.
Lending policy: {json.dumps(POLICY, indent=2)}
Applicant profile: {json.dumps(applicant, indent=2)}
Calculated DTI ratio: {dti:.2f}
Proposed monthly EMI: Rs {monthly_emi:.0f}

Evaluate this loan application:
1. Risk grade (A/B/C/D) based on credit score and DTI
2. Approve, conditionally approve, or reject
3. If approved, suggest interest rate (base 12% for grade A, +2% per grade step)
4. Key risk factors and mitigants
Respond in JSON with keys: risk_grade, decision, interest_rate, approved_amount, risk_factors, conditions"""

response = model.generate_content(prompt)
    text = response.text.strip()
    start = text.find("{")
    end = text.rfind("}") + 1
    result = json.loads(text[start:end])
    result["dti_ratio"] = round(dti, 2)
    result["monthly_emi"] = round(monthly_emi)
    return result

# Test application
applicant = {
    "name": "Amit Verma",
    "city": "Hyderabad",
    "monthly_income": 60000,
    "employment_type": "salaried",
    "credit_score": 720,
    "existing_emi": 8000,
    "requested_amount": 150000,
    "tenure_months": 24,
    "purpose": "Laptop purchase for work"
}

result = underwrite_loan(applicant)
print(json.dumps(result, indent=2))

It gives the following output,

{
  "risk_grade": "B",
  "decision": "approved",
  "interest_rate": "14%",
  "approved_amount": 150000,
  "risk_factors": "Moderate DTI at 0.43, stable salaried employment mitigates risk",
  "conditions": "Salary slip verification required",
  "dti_ratio": 0.43,
  "monthly_emi": 6250
}

The agent combines hard eligibility rules with Gemini's holistic reasoning to produce risk-graded decisions. In production, connect it to a credit bureau API for live credit scores, integrate with the loan management system for disbursement, and add an audit log for regulatory compliance. Route grade D applications automatically to a senior underwriter for manual review.

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