The Asynchronous Bureaucracy Buster

An autonomous background AI agent built on Gemini 3.5 Flash & Google Cloud that parses medical denial letters, researches healthcare laws, drafts legal appeals, and manages claims asynchronously.

Comment

Inspiration

Disputing denied insurance claims, unexpected out-of-network emergency room bills, and arbitrary specialty medication denials is a confusing, multi-step, weeks-long nightmare. Most AI tools today are static chatbots that wait for user prompts. We wanted to build an autonomous agent that takes real action—handling the heavy lifting of bureaucratic healthcare disputes in the background while you focus on life.

What it does

The Asynchronous Bureaucracy Buster is an autonomous background agent that:

  1. Ingests & Analyzes Medical Bills: Parses scanned PDFs or images to extract structured parameters (Claim ID, Denial Reason, Billed Amount, Member ID).
  2. Researches Healthcare Mandates: Automatically queries an integrated database of federal and state laws (such as the No Surprises Act 45 CFR § 149.110, ERISA § 503, and State Step-Therapy Exception Laws) to establish binding grounds for overturning the denial.
  3. Generates & Transmits Legal Appeals: Drafts formal, legally backed Level 1 Appeal letters and dispatches them to carrier grievance departments.
  4. Asynchronous Case Management (Memory Bank): Operates over days and weeks, evaluating incoming carrier responses, updating case state machines, and notifying the user only when specific input is required.

How we built it

  • AI Reasoning Engine: Gemini 3.5 Flash model integrated via the Google GenAI SDK (@google/genai).
  • Backend & State Machine: Express.js REST API with a persistent cross-session Memory Bank tracking audit trails, reasoning chains, and communication logs.
  • Frontend Dashboard: Vite + React + Vanilla CSS glassmorphic dark-mode interface with live status telemetry and an interactive GCP architecture visualizer.
  • Google Cloud Infrastructure: Containerized with Docker and configured for serverless execution on Google Cloud Run with Firestore state persistence.

Challenges we ran into

Extracting strict structured parameters from unstructured medical denial text across varying formats was challenging. We solved this by using Gemini 3.5 Flash's structured JSON output mode combined with fallback heuristic auditors. Additionally, designing an intuitive async event simulator to demonstrate weeks of long-running background operations in a short demo required careful state machine modeling.

Accomplishments that we're proud of

  • Building an end-to-end autonomous agent that performs real-world, high-utility actions instead of simple text conversation.
  • Integrating federal statutory protections directly into the model's reasoning loop to generate bulletproof appeal letters.
  • Maintaining persistent cross-session audit trails that record every decision point transparently.

What we learned

Building next-generation agents requires shifting from synchronous prompt-response loops to event-driven background architectures. Gemini 3.5 Flash's fast reasoning and structured outputs make it ideal for autonomous long-running enterprise and consumer agents.

What's next for The Asynchronous Bureaucracy Buster

  • Direct integration with Gmail and Outlook APIs for automated inbox monitoring and instant carrier response parsing.
  • Expansion to automotive insurance, credit bureau dispute claims, and utility billing disputes.
  • Integration with Google Cloud Secret Manager for zero-trust credential handling.

Built With

Share this project:

Updates

Submission history