PriorAuth Copilot — Intelligent PA Generator

Inspiration

I've seen firsthand how prior authorization slows everything down in healthcare. A doctor knows what their patient needs. The patient is waiting. But before anything happens, someone has to spend hours filling out forms, digging through policy documents, and hoping the payer approves it in time. That's not a technology problem — it's a workflow problem that AI is uniquely positioned to fix.

What frustrated me most is that this isn't a hard problem intellectually. The information exists in the patient's chart. The payer's criteria exist in policy documents. Connecting them just takes time — time that clinicians don't have. I wanted to build something that closes that gap instantly.

What It Does

PriorAuth Copilot takes a patient's clinical context and generates a complete, submission-ready Prior Authorization letter in under a minute.

It looks at the patient's diagnoses, current medications, lab results, and treatment history. It checks that against the payer's actual policy criteria. Then it writes the letter — with the right ICD-10 codes, CPT codes, medical necessity argument, and provider information already filled in.

It also tells you upfront whether the request is likely to be approved, needs more supporting documentation, or is at risk of denial — so the clinician can strengthen the case before submitting rather than waiting days for a rejection.

How I Built It

I built this entirely on the Prompt Opinion platform using Option 2 — the A2A agent path. No custom code. The platform handled the infrastructure so I could focus entirely on the clinical use case.

I connected it to the BCBS IL PPO Prior Authorization policy library, which covers high-cost procedures including MRI, PET scans, Humira, and Ozempic. I configured the SHARP context variables so the agent automatically pulls live patient FHIR data into every interaction. I enabled A2A availability so other agents on the platform can consult PriorAuth Copilot as a specialist within larger workflows.

The stack is MCP for tool access, A2A for agent communication, and FHIR for patient data — exactly what this hackathon set out to demonstrate.

Challenges

Getting the system prompt right took iteration. The policy documents are large and detailed, and I needed the agent to reason over them precisely rather than summarizing loosely. I also had to think carefully about what a real payer reviewer actually needs to see in a PA letter — this isn't just a summary, it's a legal document that needs to meet specific criteria.

What I Learned

Honestly, the Prompt Opinion platform surprised me. I expected to hit walls around FHIR token handling and A2A wiring — those are usually the hardest parts. But the platform absorbs all of that. What's left is the actual problem-solving, which is where the real value is anyway.

I also learned that prior authorization is just the entry point. The same pattern — agent pulls patient context, reasons over policy, generates a clinical document — applies to discharge summaries, referral letters, and care gap notifications.

What's Next

The immediate next step is expanding beyond BCBS IL PPO to cover UnitedHealthcare, Aetna, and Cigna policies. After that, I want to connect PriorAuth Copilot to a Clinical Evidence Agent so it can pull supporting literature automatically when the medical necessity case needs strengthening. The end goal is zero PA paperwork for clinicians — they describe what the patient needs, and the system handles everything else.

Built With

• a2a-protocol
• bcbs-il-ppo
• fhir
• genai
• hl7
• mcp
• prompt-opinion
• sharp-context