All work

10 · fintech · ai agents

AI underwriting agent

An independent AI second opinion on every loan application, with every rule outcome and tool call logged.

  • Azure OpenAI
  • OpenAI API
  • LangChain.js
  • LangGraph
  • Node.js
  • Plaid
Client
US consumer lender
Role
AI Engineer
Period
Devinware LLC · ongoing

Problem

A US consumer lender wanted an independent second assessment of each loan application next to its in-house underwriting engine, without handing approval decisions to an AI.

The assessment had to work from the same evidence underwriters use: bank data, credit and bank-behaviour reports, paystubs and fraud signals.

How it works

diagram · How an application is reviewed
  1. 01ApplicationA new loan application enters review.
  2. 02Gather evidenceTool calls pull bank data, credit and bank-behaviour reports, paystubs and fraud signals.
  3. 03Walk the policyAgents check the lending policy rule by rule.
  4. 04Log everythingEvery rule outcome and tool call is recorded.
  5. 05Second opinionThe assessment sits next to the in-house decision.

What I built

  • Tool-calling agents on Azure OpenAI that pull bank data, credit and bank-behaviour reports, paystubs and fraud signals, then walk the lending policy rule by rule.
  • Full logging of every rule outcome and tool call, so each AI assessment can be audited next to the recorded decision.
  • Durable state, with Azure OpenAI or OpenAI selected by configuration and retries when a provider rate-limits.
  • A dashboard that compares AI and recorded decisions loan by loan.

Guardrails

  • The AI never approves or denies on its own.
  • New prompts run champion vs. challenger on live applications before they replace the current one.

How it was verified

  • Evaluation baselines run in CI, so a prompt or model change can’t silently degrade decisions.
  • Models were benchmarked on historical loan files, including GPT-5.x and Grok models, to pick the best cost and quality balance.

Result

Handles hundreds of applications a day as a shadow reviewer.

Underwriters get an independent, fully logged assessment for every application.

media note

Client systems are under NDA, so this case study uses words and a diagram instead of screenshots.