-
-
End to End Architecture
Inspiration
Physicians spend up to 15 hours every week trapped in administrative paperwork, specifically navigating complex prior-authorization claims. Patients often wait days—or even weeks—for critical treatments like MRIs, diagnostic scans, and specialized medications simply because a minor clinical note or prior conservative therapy history was missed in the initial submission. We built OmniClaim AI to eliminate this friction entirely, shifting healthcare from manual form-filling to autonomous background orchestration.
What It Does
OmniClaim AI is an enterprise-grade autonomous agent running on Google Cloud that automates the end-to-end prior-authorization pipeline: EHR Integration: Ingests medical orders and autonomously extracts structured patient history and clinical notes. Policy & ICD-10 Compliance: Cross-references the requested procedure against dynamic insurance policy rulebooks to verify required prerequisites (e.g., verifying 6 weeks of documented physical therapy before approving a lumbar spine MRI). Automated Package Generation: Compiles and populates standardized prior-authorization PDF forms and API payloads. Zero-Trust Audit Trail: Logs full reasoning chains, decision paths, and state updates securely to persistent cloud memory.
How We Built It
Core Intelligence & Framework: Built using Gemini 3.5 Flash orchestrated via Google's Agent Development Kit (ADK) for multi-step reasoning and tool execution. Serverless Agent Runtime: Containerized and deployed to Google Cloud Run to execute asynchronously in the background upon receiving medical order webhooks. State & Memory Management: Utilized Google Cloud Firestore as a persistent Memory Bank storing clinical schemas, insurance policy rulebooks, and audit logs.
Challenges We Faced
Mapping unstructured clinical narrative notes into deterministic ICD-10 and CPT policy requirements was our biggest hurdle. We solved this by implementing strict function-calling schema constraints within Google ADK, ensuring Gemini 3.5 Flash accurately extracts clinical prerequisites without hallucinating medical facts.
Accomplishments That We're Proud Of
- Achieved an end-to-end prior-authorization execution time of under 5 seconds.
- Built a production-minded architecture that strictly separates agent reasoning from execution state, keeping cloud compute costs near zero.
- Proved that autonomous AI agents can actively solve high-friction enterprise problems beyond basic chat loops.
What We Learned
We learned how to leverage Google's Agent Development Kit (ADK) to build stateful, multi-tool agents that handle complex business logic reliably. We also mastered structuring zero-trust memory banks in Firestore to keep sensitive data compliant and traceable.
What's Next for OmniClaim AI
Live EHR Integration: Expanding from mock FHIR APIs to native Epic and Cerner EHR integrations. Real-time Payer Gateway: Connecting directly to real-time EDI 278 insurance submission clearinghouses.
What's next for OmniClaim AI
Live EHR Integration: Expanding from mock FHIR APIs to native Epic and Cerner EHR integrations. Real-time Payer Gateway: Connecting directly to real-time EDI 278 insurance submission clearinghouses.
Log in or sign up for Devpost to join the conversation.