MedLens 🩺📄

AI-assisted medical report explainer for patients

MedLens is a full-stack web app that helps patients understand their own medical documents.
You can securely upload reports (PDFs or images), and MedLens will:

• Run OCR on scanned documents
• Extract key medical entities (diagnoses, medications, lab values, etc.)
• Generate a plain-language summary at ~6th-grade reading level
• Show clickable citations that map each summary sentence back to the original report text

---

🌐 Live Demo

👉 Try MedLens here:
https://med-lens-six.vercel.app/sign-in

You can create an account and upload your own test documents.

---

✨ Features

• Secure authentication

  • Email + password login with reset flow
  • Social login (Google) via BetterAuth

• Document ingestion

  • Upload PDFs and images
  • OCR pipeline for scanned documents
  • Text normalization & sentence splitting

• Medical NLP

  • Entity extraction (diagnoses, medications, procedures, etc.)
  • Entity context flags like present / negated / uncertain
  • Summarization optimized for patient understanding (no heavy jargon)

• Traceable explanations

  • Summary sentences are linked back to the original report sentences
  • Citation panel explains “this summary sentence comes from these parts of the report”

• Job queue + admin tools

  • Multi-stage background jobs (OCR → sentences → entities → summary)
  • Error tracking for failed jobs
  • Admin view to inspect / clean up problematic jobs

---

🧱 Tech Stack

• Frontend & Backend: Next.js (App Router), React, TypeScript
• Styling: Tailwind CSS + custom components
• Auth: BetterAuth with email/password + Google OAuth
• Database: PostgreSQL (local or hosted, e.g. Neon) via Drizzle ORM
• AI / NLP:
  • OCR worker (Tesseract-style) for PDFs and images
  • Token classification / NER for medical entities
  • LLM summarization tuned for patient-friendly output
• Other:
  • API routes & server actions for document processing
  • Strict TypeScript types for documents, entities, jobs, and summaries

---

🖼️ Core Concepts

MedLens revolves around a few key data models:

• Document – The uploaded file (PDF or image) + basic metadata
• Sentences – The document text split into numbered sentences
• Entities – Extracted medical entities with labels and character spans
• Summary – A structured summary with:
  • summary: final text
  • questions: suggested questions for the patient to ask their doctor
  • citations: mapping from summary sentences → source sentence indexes
• Jobs – Background tasks that move a document through the pipeline

This design makes the app traceable, debuggable, and easy to extend with new AI steps.

---

🗂️ Project Structure (High Level)

.
├─ app/                 # Next.js App Router pages & routes
│  ├─ (auth)/           # Sign-in / sign-up / reset password screens
│  ├─ (root)/           # Main app views (dashboard, documents, admin, etc.)
│  └─ api/              # Route handlers (upload, jobs, etc.)
├─ components/          # Reusable UI components (layout, forms, document views)
├─ constants/           # Shared constants / enums (routes, statuses, etc.)
├─ database/            # Drizzle schema & database config
├─ lib/                 # Utility functions, AI pipeline, OCR, job helpers
├─ migrations/          # SQL migrations generated by Drizzle
├─ public/              # Static assets (logo, icons, favicons, etc.)
└─ types/               # Shared TypeScript types (Document, Entity, Summary, Job...)