How to Build a loan approval Agent Using CrewAI in TypeScript for pension funds

A loan approval agent for pension funds takes a borrower’s application, checks it against policy, validates risk and compliance constraints, and produces a recommendation with an audit trail. For pension funds, this matters because lending decisions are not just about credit quality; they must also respect fiduciary duty, member protection rules, data residency, and strict approval governance.

Architecture

• •

  Application intake service

  • •Receives borrower data, loan purpose, collateral details, and required documents.
  • •Normalizes input into a structured payload the agent can inspect.
• •

  Policy retrieval layer

  • •Pulls fund-specific lending rules, risk appetite limits, and jurisdiction constraints.
  • •Keeps policy outside the prompt so compliance can change without code changes.
• •

  CrewAI agent layer

  • •Uses a small set of specialized agents:
    • •Underwriter agent for credit and affordability analysis
    • •Compliance agent for pension-fund rules and regulatory checks
    • •Risk reviewer agent for concentration, tenor, and exposure analysis
• •

  Tooling layer

  • •Connects agents to internal systems:
    • •Credit bureau lookup
    • •KYC/AML status
    • •Document verification
    • •Portfolio exposure service
• •

  Decision and audit store

  • •Persists the recommendation, reasoning summary, tool outputs, timestamps, and policy version.
  • •Required for internal audit and regulator review.

Implementation

1) Install dependencies and define your data model

Use the TypeScript CrewAI SDK plus a schema library for request validation. Keep the request contract explicit; pension fund workflows break when inputs are loosely typed.

npm install @crewai/typescript zod dotenv

import { z } from "zod";

export const LoanApplicationSchema = z.object({
  applicantId: z.string(),
  amount: z.number().positive(),
  tenorMonths: z.number().int().positive(),
  purpose: z.string(),
  jurisdiction: z.string(),
  annualIncome: z.number().nonnegative(),
  existingDebtService: z.number().nonnegative(),
  collateralValue: z.number().nonnegative().optional(),
  kycStatus: z.enum(["verified", "pending", "failed"]),
  amlStatus: z.enum(["clear", "review", "blocked"]),
});

export type LoanApplication = z.infer<typeof LoanApplicationSchema>;

2) Create tools for policy and risk checks

The agent should not guess policy. Give it tools that return deterministic facts from your internal systems.

import { Tool } from "@crewai/typescript";

export const getPolicyTool = new Tool({
  name: "get_pension_fund_lending_policy",
  description: "Fetches current lending policy by jurisdiction and product type.",
  async execute(input: { jurisdiction: string }) {
    // Replace with real policy service call
    return {
      maxTenorMonths: input.jurisdiction === "ZA" ? 60 : 48,
      maxAmountMultipleOfIncome: 8,
      requireKycVerified: true,
      requireAmlClear: true,
      prohibitedPurposes: ["speculation", "crypto_trading"],
      dataResidency: input.jurisdiction === "ZA" ? "za" : "eu",
      policyVersion: "2026.01",
    };
  },
});

export const getExposureTool = new Tool({
  name: "get_portfolio_exposure",
  description: "Returns current exposure by sector and borrower.",
  async execute(input: { applicantId: string }) {
    return {
      currentExposureAmount: 125000,
      concentrationLimitBreached: false,
    };
  },
});

3) Wire up CrewAI agents and tasks

Use separate agents for underwriting and compliance. That keeps prompts narrow and makes audits easier.

import { Agent, Task, Crew } from "@crewai/typescript";
import { getPolicyTool, getExposureTool } from "./tools";
import { LoanApplicationSchema } from "./schema";

const underwriter = new Agent({
  role: "Senior Underwriter",
  goal: "Assess affordability, repayment capacity, and collateral strength.",
  backstory:
    "You underwrite consumer and secured loans for regulated pension fund lending programs.",
});

const complianceOfficer = new Agent({
  role: "Compliance Officer",
  goal: "Check the application against pension fund policy and regulatory constraints.",
  backstory:
    "You enforce pension fund lending rules, data residency requirements, and auditability.",
});

underwriter.tools = [getExposureTool];
complianceOfficer.tools = [getPolicyTool];

export function buildLoanApprovalCrew(rawInput: unknown) {
  const application = LoanApplicationSchema.parse(rawInput);

  const underwritingTask = new Task({
    description:
      `Analyze this loan application for affordability and risk:\n${JSON.stringify(application)}`,
    expectedOutput:
      "A structured underwriting assessment with approve/reject recommendation.",
    agent: underwriter,
  });

  const complianceTask = new Task({
    description:
      `Validate this loan application against pension fund policy:\n${JSON.stringify(application)}`,
    expectedOutput:
      "A compliance assessment with any rule breaches or approval conditions.",
    agent: complianceOfficer,
    context: [underwritingTask],
  });

  return new Crew({
    agents: [underwriter, complianceOfficer],
    tasks: [underwritingTask, complianceTask],
    verbose: true,
    process_type: "sequential",
  });
}

4) Run the crew and persist an auditable decision

Do not let the agent directly approve money movement. Return a recommendation that a workflow engine or human approver can finalize.

import dotenv from "dotenv";
dotenv.config();

async function main() {
  const crew = buildLoanApprovalCrew({
    applicantId: "APP-10291",
    amount: 250000,
    tenorMonths: 48,
    purpose: "home_improvement",
    jurisdiction: "ZA",
    annualIncome: 520000,
    existingDebtService: 90000,
    collateralValue": undefined,
    kycStatus": "verified",
    amlStatus": "clear"
}
``

---

## Keep learning

- [The complete AI Agents Roadmap](/blog/ai-agents-roadmap-2026) — my full 8-step breakdown
- [Free: The AI Agent Starter Kit](/starter-kit) — PDF checklist + starter code
- [Work with me](/contact) — I build AI for banks and insurance companies

*By Cyprian Aarons, AI Consultant at [Topiax](https://topiax.xyz).*