🩺 DoccyAI

Multimodal Medical Assistant for Preliminary Health Assessments

Because getting a first opinion shouldn't require a waiting room.

---

📖 Overview

DoccyAI is a cross-platform mobile application that acts as a preliminary health assessment companion. Users describe their symptoms through a conversational interface, and DoccyAI uses a trained ML model to suggest possible conditions — all without a single phone call to a clinic.

⚠️ Disclaimer: DoccyAI is designed for preliminary, informational purposes only. It is not a substitute for professional medical advice, diagnosis, or treatment. Always consult a qualified healthcare provider.

The project bridges mobile-first UX (Flutter) with a Python ML backend (FastAPI + scikit-learn), packaged in Docker and deployed on HuggingFace Spaces.

---

✨ Features

• 🔍 Symptom-based Disease Prediction — Logistic Regression model trained on curated symptom–disease datasets
• 📱 Cross-platform Mobile App — Flutter frontend supporting Android & iOS from a single codebase
• ⚡ REST API Backend — FastAPI backend with clean, auto-documented endpoints (/docs)
• 🐳 Fully Containerised — Docker setup for consistent local and production environments
• 🌐 Cloud Deployed — Live inference API hosted on HuggingFace Spaces
• 🔄 CORS-Configured — Secure cross-origin handling between the Flutter app and API

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│                    Flutter App (Dart)                   │
│         Cross-platform Mobile UI (Android / iOS)        │
└────────────────────────┬────────────────────────────────┘
                         │ HTTP REST (JSON)
                         ▼
┌─────────────────────────────────────────────────────────┐
│              FastAPI Backend (Python)                   │
│     /predict  →  Logistic Regression (.pkl model)       │
│     Auto-docs at /docs  |  CORS enabled                 │
└────────────────────────┬────────────────────────────────┘
                         │
            ┌────────────┴────────────┐
            ▼                         ▼
     🐳 Docker Container     🤗 HuggingFace Spaces
     (Local Dev / Prod)       (Live Deployment)

---

🧠 ML Model

Property	Details
Algorithm	Logistic Regression (scikit-learn)
Input	Binary symptom vector
Output	Predicted disease label + confidence
Serialisation	.pkl (Git LFS tracked)
Library Version	scikit-learn == 1.5.2 (pinned)

Why Logistic Regression? It's interpretable, fast at inference, and performs surprisingly well on structured symptom-disease datasets — a conscious choice of a model that a doctor could actually reason about, rather than a black-box neural net.

---

🚀 Getting Started

Prerequisites

• Flutter SDK ≥ 3.x
• Python 3.9+
• Docker (optional, for containerised setup)
• Git LFS (for .pkl model files)

git lfs install
git clone https://github.com/AsMetOP/DoccyAI.git
cd DoccyAI

---

🐍 Backend Setup (FastAPI)

cd backend

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Start the development server
uvicorn main:app --reload --host 0.0.0.0 --port 8000

API will be live at http://localhost:8000 Interactive docs at http://localhost:8000/docs

---

🐳 Docker Setup

cd backend
docker build -t doccyai-backend .
docker run -p 8000:8000 doccyai-backend

---

📱 Flutter App Setup

cd frontend

# Fetch dependencies
flutter pub get

# Run on a connected device or emulator
flutter run

Update the API base URL in the app config to point to your local backend or the live HuggingFace endpoint.

---

📡 API Reference

POST /predict

Accepts a list of symptoms and returns the predicted condition.

Request Body:

{
  "symptoms": ["headache", "fever", "fatigue"]
}

Response:

{
  "prediction": "Influenza",
  "confidence": 0.87
}

---

📂 Project Structure

DoccyAI/
├── backend/
│   ├── main.py               # FastAPI app & CORS config
│   ├── model/
│   │   └── model.pkl         # Trained LR model (Git LFS)
│   ├── requirements.txt      # Pinned dependencies
│   └── Dockerfile
│
├── frontend/
│   ├── lib/
│   │   ├── main.dart         # App entry point
│   │   ├── screens/          # UI screens
│   │   └── services/         # API service layer
│   └── pubspec.yaml
│
└── README.md

---

👥 Team

Contributor	Role
Asmet Sahoo	ML Engineer — Model, FastAPI Backend, Docker, HuggingFace Deployment
Astha Upadhyay	Database
Krish Agrawal	Testing
Sneha Das	Frontend Development

---

🛠️ Tech Stack

Layer	Technology
Mobile Frontend	Flutter / Dart
ML Backend	FastAPI + scikit-learn
ML Model	Logistic Regression
Containerisation	Docker
Cloud Deployment	HuggingFace Spaces
Version Control	Git + Git LFS

---

🔮 Roadmap

• Add multimodal input (image + symptom text)
• Expand disease coverage with a richer dataset
• Add explainability layer (SHAP values per prediction)
• Integrate appointment booking / doctor referral
• Offline inference with on-device model (TFLite / ONNX)

---

📄 License

This project is licensed under the MIT License.

---

_{Built with ❤️ by the DoccyAI team · KIIT University, Bhubaneswar}