Inspiration
Prior authorization is one of healthcare’s clearest operational pain points. Clinicians and revenue cycle teams must manually search records, interpret payer criteria, draft medical necessity language, and coordinate follow-up. AuthBridge AI was built to show how healthcare agents can move beyond chat and produce auditable workflow outputs.
What it does
AuthBridge AI prepares a human-reviewable prior authorization readiness packet for a synthetic advanced imaging request. It maps patient evidence against payer-style criteria, identifies missing or weak evidence, drafts a medical necessity rationale, produces a packet-readiness status, and creates a follow-up task. It does not approve or deny care.
How we built it
We built AuthBridge AI inside Prompt Opinion as a patient-scope A2A-enabled agent with FHIR Context Extension support. The demo uses existing synthetic patient context in the Prompt Opinion workspace plus a synthetic prior authorization request and synthetic payer-style criteria. The agent produces workflow outputs aligned to Talk, Table, Template, Transaction, and Task: a summary, criteria-evidence table, draft rationale, readiness status, and follow-up task.
Challenges we ran into
The main challenge was keeping the solution narrow, safe, and feasible under the hackathon timeline. We avoided building a generic healthcare chatbot and focused on a specific operational workflow: prior authorization readiness. We also designed the agent to avoid unsafe approval or denial claims, avoid invented facts, and require human review before submission.
Accomplishments that we're proud of
We created a working Prompt Opinion marketplace agent that demonstrates how generative AI can assemble an evidence-mapped prior authorization packet from synthetic patient context and payer-style criteria. The output is operationally useful, human-reviewable, and structured for healthcare workflow handoff.
What we learned
The strongest healthcare agent workflows combine patient context, standards-based interoperability, structured outputs, and human governance. FHIR provides patient context, A2A enables interoperable agent invocation, and generative AI adds value by synthesizing mixed evidence into workflow-ready artifacts.
What's next for AuthBridge AI
Next steps would include connecting to real scoped FHIR sandbox data, adding payer-specific policy retrieval, supporting additional prior authorization categories, integrating task routing for revenue cycle teams, and strengthening evaluation/guardrail coverage. A production version would require privacy, security, audit, payer-policy validation, and clinical governance review.
Log in or sign up for Devpost to join the conversation.