MediEase

A doctor in your pocket. MediEase turns symptoms, lab results, and medical jargon into plain-English answers and next steps, fast. Private, insights for smarter health decisions.

Comment

Inspiration

Healthcare info is confusing and urgent. Friends and family struggle to decode lab reports and discharge notes. We wanted a calm, private “explain it to me simply” layer that guides next steps without replacing a doctor.

What it does

  • Turns symptoms, lab results, prescriptions, and reports into plain-English summaries, medical interpretation, red-flag checks, and next-step guidance.
  • Supports paste, upload (PDF/images), with voice coming soon.
  • Adds confidence cues, sources, and a copy/print/PDF export for sharing with a clinician.
  • Privacy-first: local preprocessing + minimal retention; clear disclaimers.

How we built it

  • Frontend: React + TypeScript (Vite) with Tailwind and shadcn/ui for fast, accessible UI.
  • AI Layer: Supabase Edge Function (TypeScript on Deno) orchestrates GPT-4 with strict JSON schemas + function calling.
  • Data/Auth: Supabase for auth, storage (PDFs), and optional Postgres/pgvector to cache vetted clinical references.

Challenges we ran into

  • Keeping explanations accurate and non-alarmist while avoiding hallucinations.
  • Normalizing messy PDFs and inconsistent lab units/ranges.
  • Balancing model quality vs. speed and cost on an edge runtime.

Accomplishments that we're proud of

  • End-to-end flow from paste/upload → structured analysis → one-click PDF export.
  • Median response near ~6s with graceful streaming UI.
  • Guardrails + schema that keep outputs predictable and clinician-friendly.

What we learned

  • In healthcare, trust > features: tone, transparency, and safety rails matter most.
  • Strict JSON contracts + retrieval drastically improve consistency.
  • Small UX details (confidence chips, expandable sections) reduce anxiety and boost comprehension.

What's next for MediEase

  • Live Doctor Connect
  • Image uploading
  • Medication checker and interaction warnings.
  • Multilingual support and lightweight mobile app.
  • Ongoing clinical advisory reviews, evaluation datasets, and stronger guardrails for edge cases.

Built With

Share this project:

Updates