Sadam Khan

AI · Healthcare

Medical Imaging AI

Diagnostics with fine-tuned medical models.

Role
Full-Stack AI Developer
Period
2025

Outcomes

87% accuracy on X-ray fracture detection
84% accuracy in ophthalmology; 23 dermatology conditions classified
Automated report pipeline processing 15K+ images

The problem

Clinics needed faster preliminary reads across radiology, ophthalmology, and dermatology without shipping patient data to opaque third-party services.

The approach

Fine-tuned MedGemma 4B for X-ray fracture detection, ophthalmology, and 23 dermatology conditions; curated a 30K+ medical image dataset; built an automated imaging report pipeline (15K+ images processed) and deployed distributed GPU inference behind FastAPI, with web and React Native clients under HIPAA/FHIR standards.

Stack

  • Next.js
  • FastAPI
  • Django
  • PyTorch
  • MedGemma
  • GPU Computing
  • React Native
  • PostgreSQL
  • AWS Lambda
Medical Imaging AI system architecture diagram
System architecture. The product interface is not public, so this diagram shows how the platform fits together.
MedGemma model card on Hugging Face
MedGemma, the open medical model the platform fine-tunes
← All workNext: SiteScripter AI →