Description
Summary We run a lending platform built on PHP and want to add an AI-assisted underwriting review step to our existing loan review workflow. Important framing: this is decision support, not automated decisioning. The AI produces a structured risk summary; our human underwriters make every credit decision. Scope (~10 hours): A PHP service class that sends structured application data (income, employment, requested amount, DTI inputs, history flags) to the OpenAI or Claude API Returns validated JSON: risk flags, inconsistencies, suggested stipulations, and a short reviewer memo Render that output in our existing underwriter review screen (one view/component, we'll point you to it) Prompt template in config, JSON schema validation with retry on malformed output Log all AI inputs and outputs to our audit table Out of scope: document OCR/parsing, credit bureau integration, model training, and any auto-approve/deny logic. Stack: PHP 8.x, MySQL, Guzzle for the API calls. You'll work in a feature branch against staging with seeded test applications. Budget: $70–90/hr. Estimated ~10 hours, weekly limit set at 12 hours. Timeline: one week. To apply: briefly describe an LLM integration you've done in PHP (or a similar backend) and how you handled malformed model output. Generic proposals will be skipped.