🚀 Welcome! Ready to streamline your Prior Authorization process? Click below to get started with your deployment and unlock the full potential of AutoAuth.

• Overview
• Introducing AutoAuth
• Quick Start
  • End-to-End Deployment Using AZD
  • PriorAuth SDK
  • Evaluations Framework
• Responsible AI
• What's Next?
• Contributors & License

Prior Authorization (PA) is a critical step in healthcare delivery, requiring providers to seek approval from payors before offering certain treatments. While essential for cost control and care management, the current PA process is often manual, fragmented, and time-consuming:

• Provider Burden: Physicians handle an average of 41 PA requests per week, consuming 13 hours—almost two full working days—leading to high administrative burdens (88% of physicians report it as high or extremely high). [1]
• Payor Costs: Up to 75% of PA tasks are manual, costing around $3.14 per request, and can be reduced by up to 40% through AI-driven automation. [2] [3]
• Patient Outcomes: 93% of physicians state PA delays necessary care, and 82% of patients sometimes abandon treatments due to these delays. Even a one-week delay in critical treatments like cancer can increase mortality risk by 1.2–3.2%. [1] [3]

This repository aims to streamline and automate the PA process using Azure AI, Agentic workflows, and advanced reasoning models. By leveraging machine learning, OCR, and agentic retrieval-augmented generation (RAG), we can reduce human labor, cut costs, and ultimately improve patient care.

Tackling time-heavy tasks within the Prior Authorization workflow.

Note: Our methodology, developed in collaboration with clinicals experts (MD and PharmD), is research-based and includes comprehensive case studies. For an in-depth understanding, please refer to our article on Hashnode..

AutoAuth revolutionizes the Prior Authorization process through:

• Intelligent Document Analysis: OCR and LLM-driven extraction of clinical details from various document types.
• Smart Policy Matching: Agentic Rag laveraging Hybrid retrieval systems (Vector + BM25) identify relevant policies and criteria swiftly.
• Advanced Reasoning Models: Assess compliance against policies, recommend Approve/Deny decisions, or request additional info with full traceability.

Solution Architecture

Note: For comprehensive details, including technical architecture, customization steps, references, and additional documentation, please visit our GitHub Pages.

Click the image to watch the AutoAuth App demo.

More detailed documentation can be found in docs/azd_deployment.md.

You can seamlessly integrate Prior Authorization (PA) processing into your application using our SDK. The SDK allows you to run PA workflows programmatically, enabling you to automate the end-to-end process.

from src.pipeline.paprocessing.run import PAProcessingPipeline

# Instantiate the PA processing pipeline
pa_pipeline = PAProcessingPipeline(send_cloud_logs=True)

# Run the pipeline with uploaded files
await pa_pipeline.run(uploaded_files=files, use_o1=True)

For those looking for greater flexibility, the AutoAuth SDK enables you to embed PA microservices into your existing applications. You can customize, integrate, and extend the PA workflows to suit your specific needs. This approach provides a highly modular, code-first experience for developers who want to build their own solutions.

• 📡 Plug-and-Play API Integration with FastAPI: Quickly expose Prior Authorization (PA) workflows as REST APIs, enabling system-to-system integrations.
• 🔄 Modular and Extensible for Custom PA Workflows: Customize and extend the SDK to fit your business logic and workflows.
• ⚡ Rapid Deployment and Integration: Minimal setup required to start automating PA workflows. Use FastAPI or other framework to expose endpoints and interact with the PA logic programmatically.

With the AutoAuth SDK, you have the flexibility to automate end-to-end Prior Authorization workflows or select specific components to integrate into your system. Whether you require a full application or a microservice solution, AutoAuth provides the tools you need.

Note: Detailed information, technical architecture, customization steps, references, and further documentation are available on our GitHub Pages.

For those looking to understand how quality of decisions and rationale changes, automated tests have been implemented as part of this repository in a declarative manner. You can use these, extend them or further enhance them to suit your needs, using the same pipeline components that are used to deliver the end-to-end PA pipeline.

• ⚙️ AutoAuth Evaluation Framework: Leverages a configuration-driven approach to define and execute evaluation cases for generative AI tasks. Test configurations are decoupled from implementation code, enhancing clarity and maintainability.
• 🧩 Separation of Concerns: Uses dedicated Python modules (with evaluator.py) to execute evaluations defined via YAML.
• 🚀 Flexibility & Scalability: Easily update evaluation parameters, add new evaluators, or integrate additional test cases without modifying core pipeline logic.
• 🔄 Standardized Workflow: Employs a robust three-step process—preprocessing, running evaluations, and post-processing—to ensure consistent execution alongside the Prior Auth pipeline.
• 🏭 AI Foundry Integration: Seamlessly catalogs and monitors performance metrics for each evaluation case, with support for both integrated and custom evaluators.

For a detailed explanation, please see our Evaluations Documentation.

To get started viewing evaluation runs in AI Foundry:

azd up
pip install -r requirements.txt
pytest

This project was designed with Responsible AI (RAI) principles in mind from day one:

• Bias Reduction: The system is not designed to make final decisions (e.g., approve or deny). Instead, it acts as a decision-support tool that surfaces facts by cross-referencing structured clinical data. This approach helps reduce bias, eliminate subjectivity, and ensure consistent outputs (focus on facts).
• Human-in-the-Loop: AutoAuth is designed to support clinical experts—not replace them. Final determinations are always made by humans, with the AI providing transparent rationale along the way.
• Explainability & Transparency: Every decision path is traceable. Clinicians can understand what data was used, how it was interpreted, and what led to the suggested outcomes.
• Data Privacy & Governance: All demos and tests use synthetic data. Customers are expected to conduct their own RAI impact assessments before going live, ensuring proper alignment with internal policies and regulations.
• Evaluation & Monitoring Framework: We created a companion evaluation suite, MedEvals, to rigorously test the quality and rationale of PA decisions using Azure AI Foundry. This enables continuous monitoring, validation, and fairness tracking.

ℹ️ For more information, check out our MedEvals Evaluation Framework

• Agentic Framework Leveraging Semantic Kernel: Integrate the Agentic framework component using the Semantic Kernel as the core for context-aware and intelligent agent orchestration.
• API Management (APIM) Integration: Introduce APIM for secure, scalable, and controlled access to the service’s endpoints.

Please read through our contributing guidelines. Directions are included for opening issues, coding standards, and notes on development.

License: MIT License

The software may collect information about you and your use of the software and send it to Microsoft. Microsoft may use this information to provide services and improve our products and services. You may turn off the telemetry as described in the repository. There are also some features in the software that may enable you and Microsoft to collect data from users of your applications. If you use these features, you must comply with applicable law, including providing appropriate notices to users of your applications together with a copy of Microsoft’s privacy statement. Our privacy statement is located at https://go.microsoft.com/fwlink/?LinkID=824704. You can learn more about data collection and use in the help documentation and our privacy statement. Your use of the software operates as your consent to these practices.

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft’s Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party’s policies.