Advanced Clinical Documentation Assistant (ACDA)

Turn clinician-patient conversation transcripts into schema-validated SOAP notes in seconds using Google Gemini and a secure Cloud Run proxy.

Comment

Check the YouTube Video Demo at: https://www.youtube.com/watch?v=863xBiL0h1I The video is not loading for some reason.so check at YT Link.

Inspiration

Healthcare professionals in India spend Several hours daily on clinical documentation, reducing patient care time. I built ACDA to automate SOAP note generation using AI, helping clinicians focus on what matters most—their patients.

What it does

ACDA accepts free-text clinical transcripts and returns validated SOAP notes (Subjective, Objective, Assessment, Plan) in both JSON and human-readable formats. The React frontend submits transcripts to a secure Node/Express backend on Google Cloud Run, which:

  • Holds the Gemini API key in Secret Manager
  • Calls Google Gemini API
  • Enforces response schema validation
  • Returns structured clinical notes

Users can review, download PDFs (using jsPDF + html2canvas), and export notes for EHR workflows. The app includes Capacitor for Android APK builds.

How I built it

I build it using Google AI Studio

Challenges I ran into

  1. Large transcripts required chunking and schema enforcement to maintain consistency
  2. Build optimization - Vite apps in containers needed dev dependencies during build stage
  3. Security hardening - Implementing Secret Manager + IAM roles for production-grade key management
  4. Schema validation - Ensuring Gemini responses match strict SOAP JSON structure

Accomplishments that I'm proud of

  • Schema-validated output: Gemini responses validated against strict SOAP JSON for downstream EHR integration
  • Medication safety: Automatic dosage validation and interaction warnings
  • Security-first design: API keys never exposed to client, proxied through Cloud Run with Secret Manager
  • Production-ready: Full CI/CD pipeline with Cloud Build

What I learned

  • Building secure serverless AI applications requires careful API key management
  • Schema enforcement is crucial for healthcare applications
  • Cloud Run's autoscaling handles variable clinical workflow loads efficiently
  • Containerizing Vite applications requires careful dependency management

What's next for ACDA

  • Multi-tenant architecture with role-based access control
  • FHIR mapping for EHR system integration
  • Clinical validation pipeline and institutional review
  • Explainability UI and clinician feedback loop to improve prompts
  • Real-time voice transcription integration

Built With

  • cloud
  • geminisecret
  • googleaistudio
  • reactvitetypescriptnode.jsexpressgoogle
  • rungoogle
Share this project:

Updates