Format: no restrictions

The Task

Build a proof-of-concept demonstrating the GenAI document review feature integrated into the existing platform. Your code must include both backend and frontend:

Backend (Java / Spring Boot)

• A new service or module that accepts an uploaded pay stub, sends the relevant content to an LLM API for extraction, and returns structured data (employer, income, pay period, YTD totals, discrepancy flags)
• RESTful API endpoint(s) that the Angular frontend will consume
• Integration with the existing Application Service and Document Service — show how the extracted data connects to the application record
• PII handling: demonstrate how sensitive data is redacted or protected before being sent to the LLM API

Frontend (Angular)

• A dashboard component where loan officers can see the AI-extracted data alongside the applicant's self-reported data
• Visual flagging of discrepancies (e.g., income mismatch between pay stub and application)
• An approval workflow: the loan officer reviews the AI extraction, can accept, edit, or reject each extracted field before it flows to underwriting

You may use mock data and stub external service calls (S3, LLM API) where needed. The proof-of-concept should demonstrate the design, data flow, and integration points — it does not need to run against live infrastructure.

Format: .pdf, .doc, .docx, .rtf, .txt, .md

Written Document

A written document with three sections:

Section A — Architecture & Design Rationale (300–500 words)

• How does your GenAI document review service fit into the existing microservices architecture? Where does it sit relative to the Application, Document, and Underwriting services?
• What were the key design decisions and trade-offs you made? What alternatives did you consider?
• How does data flow from document upload through AI extraction to loan officer review to underwriting? Include a simple diagram if helpful.
• How did you handle the PII redaction requirement, and what are the trade-offs of your approach?

Section B — Production Readiness & Leadership (200–400 words)

• Part B1: What would need to change before this proof-of-concept goes to production? Cover at minimum: error handling for LLM failures or hallucinations, monitoring and observability, security hardening, and rollback strategy if the AI extraction produces bad results at scale.
• Part B2 — Required reasoning question (answer without AI assistance): Describe a scenario where an AI coding assistant would give you a plausible but incorrect answer for this type of problem — and explain specifically how you would catch it. What would the incorrect output look like, and what would you check to identify the error before acting on it?

Section C — AI Usage Log (Mandatory)

This is not a trick. We want to see how you work with AI — not whether you used it.

In a short section of your README, document your AI collaboration process. For each significant interaction with an AI tool, briefly note:

• What you asked the AI to help with
• What it gave you
• What you kept, changed, or rejected — and why

Three interactions documented is sufficient. The log does not need to be exhaustive.

Format: .mp4, .mov, .webm

Video Walkthrough

Record a screen-share video walking through your submission. Upload it as an MP4 or MOV file directly on the Provn platform as a separate file. Structure your video as follows:

• Summary (~1 minute): The problem and your approach
• Code walkthrough (~3–4 minutes): Walk through your code — both backend and frontend. Explain your design decisions, not just what the code does. How do the services interact? Why did you structure it this way?
• Architecture & production reasoning (~2–3 minutes): How does your service fit into the microservices architecture? What would you prioritize for production readiness? How would you mentor a junior developer working on this codebase?
• Mandatory AI question (~1–2 minutes): Walk me through one moment where you disagreed with, pushed back on, or redirected what the AI gave you — and what you did instead. Name the specific moment. Explain what the AI produced that didn't meet the bar, what you did differently, and why.
• Reflection (~30–60 seconds): What would you do differently with more time?

Speak naturally. Communication is assessed on clarity of technical ideas and logical structure — not verbal polish, accent, or filler words.