Doc Assist

Doctors spend a bunch of their time reviewing patient data which is scattered and doesn't paint the full picture. We created DocAssist to centralize patient data and let doctors tap into it using AI.

Comment

Inspiration

Doctors spend 40% of their time on paperwork instead of patients. Medical records are scattered across systems, making it hard to find relevant history quickly.

What it does

DocAssist is an AI-powered clinical documentation platform with three core features:

  • Smart Clinical Briefs: Generates complaint-focused summaries from patient records
  • RAG Chat: Answer questions about patient documents with citations
  • SOAP Note Generation: Auto-generates structured clinical notes

How we built it

  • Frontend: React, TypeScript, Vite, Tailwind CSS
  • Backend: Supabase (Auth, Database, Storage) with Row Level Security
  • AI: Lovable for interfaces, Trae for IDE, Keywords AI for LLM observability and monitoring

Challenges we ran into

  • Implementing proper data isolation so doctors only see their own patients
  • Tuning RAG retrieval to surface relevant medical context without hallucinations
  • Balancing response speed with accuracy for real-time clinical use

Accomplishments that we're proud of

  • End-to-end working prototype with secure multi-tenant architecture
  • Citation-backed AI responses that doctors can trust and verify
  • Full observability into every AI call through Keywords AI

What we learned

  • Keywords AI makes LLM debugging and iteration dramatically faster
  • Row Level Security in Supabase is powerful for healthcare-style data isolation
  • Prompt engineering for medical contexts requires extra precision

What's next for Doc Assist

  • Integration with EHR systems (Epic, Cerner)
  • Voice-to-note functionality for hands-free documentation
  • Expanded clinical decision support features

Built With

Share this project:

Updates