AI-powered multimodal disease screening platform using photo, voice, and symptom analysis. Offline, multilingual, privacy-first healthcare assistant for low-resource communities.
# π©Ί MediScan AI
**An offline-first, multimodal AI health-screening assistant for low-resource communities.**
---
## π― What It Does
MediScan AI is a **privacy-first, multimodal health-screening platform** built for communities with limited access to doctors and stable internet. A patient can describe symptoms by **photo, voice, or text** β the system classifies the issue and generates a preliminary assessment, entirely **without sending data to the cloud**.
Built as a submission to **Samsung Solve for Tomorrow**.
## β¨ Key Features
- πΌοΈ **Skin condition classification** β fine-tuned EfficientNet-B3 model for visual symptom screening
- ποΈ **Voice-based symptom intake** β Whisper transcribes spoken symptoms in multiple languages
- π§ **RAG-powered symptom analysis** β BioMistral-7B + LangChain + FAISS retrieve relevant medical context before generating an assessment, reducing hallucination
- π± **Cross-platform mobile app** β Flutter frontend designed for low-end devices and intermittent connectivity
- π **Privacy by design** β inference runs locally/offline wherever possible; no raw patient data leaves the device unnecessarily
- π **Multilingual support** β built for accessibility across non-English-speaking communities
## ποΈ Architecture
```
βββββββββββββββ ββββββββββββββββ ββββββββββββββββββββββ
β Flutter App β βββΆ β FastAPI β βββΆ β EfficientNet-B3 β (image β classification)
β (Patient UI)β β Backend β βββΆ β Whisper β (voice β text)
βββββββββββββββ β β βββΆ β BioMistral-7B β (RAG symptom analysis)
ββββββββββββββββ β + LangChain + FAISS β
ββββββββββββββββββββββ
```
## π οΈ Tech Stack
| Layer | Technology |
|---|---|
| Backend | FastAPI (Python) |
| Image Model | PyTorch, EfficientNet-B3 |
| Speech-to-Text | OpenAI Whisper |
| Symptom Reasoning | BioMistral-7B, LangChain, FAISS (RAG) |
| Mobile Frontend | Flutter |
| License | Apache 2.0 |
## π Getting Started
```bash
# Clone the β¦