
MediScan AI is an end-to-end AI-powered medical imaging diagnostic platform built solo for the AMD Developer Hackathon, covering Track 1 and Track 3. Track 3 — Vision & Multimodal AI: Three EfficientNet-B0 CNN models were trained on real medical datasets — skin cancer detection (7 classes, 9,577 images), chest X-ray pneumonia classification (5,216 images), and eye disease detection (4,217 images). Each model delivers accurate predictions from a single image upload through a clean Gradio interface. Track 1 — AI Agents & Agentic Workflows: A LangChain-powered AI agent sits on top of the Vision AI layer to interpret CNN predictions, explain diagnoses in simple language, answer patient follow-up symptom questions, and recommend nearby doctors — making AI diagnostics truly accessible to everyone. Future Plans: Expand disease coverage to include diabetic retinopathy, tuberculosis, and brain tumors. Build a mobile app for on-the-go medical image scanning in low-resource settings. Add multi-language support for rural healthcare access across India and beyond. Built using PyTorch, EfficientNet-B0, Transformers, LangChain, Gradio, FastAPI, and HuggingFace Spaces.
10 May 2026