MediScan AI

Upload any medical report → AI reads, explains and translates it in 22 Indian languages

Comment

Inspiration

1.4 billion Indians receive medical reports they cannot understand. Most wait days for a doctor appointment just to be told what their blood test means. Rural patients, elderly citizens, and non-English speakers are most affected. We wanted to bridge this gap — giving every Indian immediate, clear understanding of their own health data in their own language.

What it does

MediScan AI lets anyone upload a medical report — blood test, lab result, X-ray, or prescription — and instantly understand it in plain language in their mother tongue.

  • Reads the report using Amazon Nova Pro's multimodal AI (PDFs and images)
  • Extracts every finding automatically with reference ranges
  • Classifies each result as 🟢 Normal / 🟠 Warning / 🔴 Critical
  • Explains results in simple language anyone can understand
  • Translates the entire analysis into any of India's 22 official languages
  • Downloads the full report as PDF

How we built it

  • AI Model: Amazon Nova Pro (amazon.nova-pro-v1:0) via Amazon Bedrock Converse API
  • Multimodal: Nova reads both PDF documents and images using document and vision understanding
  • Agentic: A single Nova call autonomously reads → extracts → classifies → explains → recommends with no manual prompting at each step
  • Translation: A separate Nova API call translates the full analysis on demand — results cached for instant switching
  • Backend: Python 3 with stdlib HTTP server + boto3
  • Frontend: Pure HTML/CSS/JS — zero frameworks, zero build steps

Challenges we ran into

  • API timeouts: Initially tried to translate into all 22 languages in one API call — this caused read timeouts. Fixed by separating analysis and translation into two endpoints, with translation triggered only when user selects a language.
  • File handling: Parsing multipart form data in pure Python stdlib without any web framework was tricky but kept dependencies minimal.
  • WSL file path issues: Development on WSL required careful path handling for serving the HTML frontend.

Accomplishments that we're proud of

  • Built a fully working AI medical analyzer in a single day
  • Successfully tested on real blood reports from Tagore Medical College, Chennai — correctly identified critical Vitamin D deficiency
  • All 22 official Indian languages supported with full UI translation
  • Entire app ru

Built With

Share this project:

Updates