RYAN ZERNACH

Full-Stack AI Systems Engineer

Ryan_Zernach_2025_Senior_AI_Systems_Engineer_Remote_United_States

BUILDING RELIABLE

HEALTHCARE AI

SCALING PRODUCTION

LLM SYSTEMS

ENGINEERING TRUSTED

AI INFRASTRUCTURE

DELIVERING INTELLIGENT

PATIENT IMPACT

🪉 Lyzr AI Auto-Loan Underwriting Copilot

I built and shipped Lyzr AI in less than 2 days: a full-stack underwriting copilot that turns loan policy and applicant data into a structured YES/NO recommendation with an explanation. Instead of asking a reviewer to interpret a generic chat response, the live product at lyzr.archlife.org brings together a React frontend, FastAPI backend, Firebase-backed data layer, and Lyzr ADK orchestrator that routes each case through credit, affordability, collateral, compliance, and escalation subagents.

Explore
Live App
GitHub Repo
Lyzr platform logo
The attached Lyzr platform logo used throughout the release.

What shipped

This was not a mockup or a one-screen demo. It was a deployable product with real request handling, persistence, and a live endpoint.

  • A React frontend that makes underwriting a focused decision cockpit rather than a generic chat surface.
  • A FastAPI backend that validates requests and hands each case cleanly to the agent layer.
  • A Firebase-backed data layer with real persistence, not just transient demo state.
  • A Lyzr ADK orchestrator that routes each case through credit, affordability, collateral, compliance, and escalation subagents.
  • A live deployment at lyzr.archlife.org and a public GitHub repo for the full implementation.

What made speed possible

Speed came from a tight problem and an explicit execution path: one job, one primary user flow, and a clean split between presentation, orchestration, and data.

  • The scope was narrow enough to ship fast, yet real enough to prove full-stack delivery.
  • The UI leads with the recommendation and explanation, so a human reviewer can use it immediately.
  • The frontend, backend, and database deployed together, making the app behave like a product rather than a prototype.
  • The live system favors structured output, traceable decisions, and clear operator feedback over prompt theater.

A stack with clear seams

The implementation stayed simple in the useful way. Each layer owned a clear responsibility, so the system could ship without turning into a tangle.

Frontend

Backend

Database

Release

From concept to a live decision surface in less than 2 days

That included the deployed frontend, backend, database, and live integration path. The result is an AI product that can be opened, reviewed, and iterated on immediately—not just described in slides.