Healthcare Superpower Agent

A2A healthcare agent with FHIR support for medication safety - built entirely on Android for Agents Assemble

Comment

Inspiration

Medication errors and dangerous drug interactions cause thousands of preventable hospitalizations each year. I wanted to build an AI agent that could help patients and clinicians quickly identify medication safety concerns - specifically the common but dangerous combination of ACE inhibitors (like Lisinopril) with NSAIDs (like Ibuprofen), which can cause acute kidney injury.

What it does

Healthcare Superpower Agent is an A2A-enabled healthcare AI that provides:

  • Medication side effect identification (e.g., Metformin's GI effects and B12 deficiency risk)
  • Drug-drug interaction checking (e.g., Lisinopril + Ibuprofen renal risk)
  • FHIR patient data parsing with SHARP extension support
  • Clinical recommendations with proper medical disclaimers

The agent responds with structured, clinically accurate information followed by a safety disclaimer.

How we built it

Platform: Prompt Opinion (A2A + MCP + FHIR) AI Model: GitHub Models (openai/gpt-4.1) Authentication: GitHub Fine-grained PAT with Models permission Testing: Launchpad chat interface on Android

Built entirely on an Android phone using Termux, Chrome browser, and GitHub.

Architecture:

  • A2A protocol enabled with FHIR Context Extension
  • Skills: medication_safety_check, parse_fhir_patient
  • Published to Prompt Opinion Marketplace

Challenges we ran into

  1. Google Cloud Vertex AI setup - Hit 403, 400, billing, and API enablement errors. Switched to GitHub Models for simpler authentication.
  2. MCP session headers - Encountered "Mcp-Session-Id required" errors; resolved by using the platform's built-in A2A interface instead of direct API calls.
  3. JSON response format error - Agent failed with "schema must be type: object"; disabled JSON format for plain text responses.
  4. Finding the test interface - The Launchpad was hidden; discovered it through the "Launch" button in BYO Agents view.

Accomplishments that we're proud of

✅ Clinically accurate medication interaction checking (tested: Metformin side effects + Lisinopril/Ibuprofen interaction) ✅ Proper medical disclaimers on every response ✅ A2A enabled with FHIR Context Extension ✅ Fully functional agent published to Marketplace ✅ Built entirely on an Android phone - no laptop required!

What we learned

  • A2A protocol enables seamless agent-to-agent communication in healthcare
  • FHIR + SHARP extensions provide critical patient context for clinical AI
  • GitHub Models offers a simpler alternative to Vertex AI for prototyping
  • Proper error handling and disclaimers are essential for healthcare AI

What's next for Healthcare Superpower Agent

  • Connect to live FHIR servers (HAPI/Firely) for real patient data
  • Add more medication interaction databases (DrugBank, RxNorm)
  • Build specialized MCP tools for cardiovascular and diabetes medications
  • Publish to the Prompt Opinion Marketplace for clinical pilot testing

Built With

  • a2a-protocol
  • fhir
  • fhir-context-extension-(sharp)
  • github-fine-grained-pat
  • github-models
  • json-rpc
  • mcp-(model-context-protocol)
  • openai/gpt-4.1
  • prompt-opinion-platform
  • termux
Share this project:

Updates