🩺 Assist AI - Medical Assistant Platform

A human-in-the-loop AI platform designed to bridge the communication gap between patients with chronic conditions and their doctors.

This project was developed for a hackathon to demonstrate how AI can empower healthcare professionals and reassure patients by acting as an intelligent, safe, and efficient intermediary. It provides a draft suggestion for patient queries which is always reviewed and approved by a doctor before being sent.

🖼️ Project Gallery

A brief look at the Patient and Doctor portals.

Patient Portal - Submit Query

Doctor Portal - Review Dashboard

(Insert screenshot of the patient query submission screen here)

(Insert screenshot of the doctor's dashboard with pending queries here)

🚀 The Problem

For millions living with chronic diseases like diabetes, timely access to their doctor for non-emergency questions is a constant challenge. A strange blood sugar reading or a question about a meal can cause significant anxiety. This "care gap" creates a communication bottleneck for doctors and leaves patients feeling uncertain. Assist AI was built to solve this problem.

✨ The Solution: How It Works

Assist AI provides a seamless and trustworthy workflow:

Patient Submits a Query: A patient logs into their portal and submits a health-related question, optionally attaching relevant documents like lab results.

AI Generates a Draft: The backend, powered by a LangGraph agent, receives the query. It fetches the patient's specific medical history and uses an LLM (DeepSeek) to generate a safe, empathetic, and informative draft response.

Doctor Reviews & Approves: This AI-generated draft is never sent directly to the patient. Instead, it appears on the doctor's secure dashboard for review. The doctor has full control to approve, edit, or completely rewrite the response.

Patient Receives Verified Advice: Only after the doctor approves the final message is it sent to the patient and logged in their history, ensuring all advice is medically sound and verified.

核心功能 (Core Features) Dual-Role Interface: Separate, secure portals for Patients and Doctors.

Intelligent Query Processing: An AI agent that considers patient history to provide context-aware suggestions.

Doctor-in-the-Loop: Ensures 100% human verification of all medical advice.

Query Management: A dashboard for doctors to review pending queries and approve responses.

Patient History: A view for patients to see their past submissions and the doctor's final responses.

🛠️ Tech Stack

Backend:

Framework: FastAPI

AI Orchestration: LangGraph

LLM: DeepSeek (via an OpenAI-compatible API)

Data Validation: Pydantic

Frontend:

Framework: Streamlit

Core Language: Python

📂 Project Structure The project is organized into two main components for a clean separation of concerns:

hackathon-demo-assist/ ├── backend_service/ # FastAPI backend, LangGraph agent, and all core logic │ ├── main.py # API endpoints │ ├── graph.py # LangGraph workflow definition │ ├── pending_queries.py # Logic for saving/loading queries │ ├── patient_db.py # Mock patient database │ └── ... └── assistMVP/ └── streamlit_app/ # All frontend Streamlit files ├── main.py # Main UI layout and role selection ├── patient.py # Patient portal UI └── doctor.py # Doctor dashboard UI

🏃 Getting Started

To run the application locally, you'll need to set up the environment and run both the backend and frontend services concurrently.

Prerequisites

Python 3.9+

An LLM API Key (e.g., from DeepSeek)

Installation & Setup

Clone the repository:

git clone https://github.com/your-username/hackathon-demo-assist.git cd hackathon-demo-assist

Set up the Backend:

Install backend dependencies:

pip install -r backend_service/requirements.txt

Create an environment file by copying the example:

cp backend_service/.env.example backend_service/.env

Edit backend_service/.env and add your DeepSeek API key and base URL.

Set up the Frontend:

Install frontend dependencies:

pip install -r assistMVP/streamlit_app/requirements.txt

Running the Application You need to open two separate terminals.

Terminal 1: Start the Backend Server

Navigate to the project root (hackathon-demo-assist) and run:

python -m uvicorn backend_service.main:app --reload --port 8000

The backend will be running on http://localhost:8000.

Terminal 2: Start the Frontend Application

Navigate to the same project root (hackathon-demo-assist) and run:

streamlit run assistMVP/streamlit_app/main.py --server.port 8502

The frontend will be accessible at http://localhost:8502.

🌐 API Endpoints

The backend provides the following key API endpoints:

Method

Endpoint

Description

POST

/process_query/

Submits a new patient query for AI processing.

GET

/pending_queries/

Fetches all queries awaiting a doctor's review.

POST

/update_query/{query_id}

Updates a query's status (e.g., to "approved").

GET

/queries/by_patient/{patient_id}

Fetches the complete history for a specific patient.

🔮 Future Work

Persistent Database: Replace the in-memory/JSON storage with a robust database like PostgreSQL or MongoDB.

Real-time Notifications: Implement WebSockets to notify doctors instantly when a new query arrives.

Authentication & Security: Add a proper user authentication system (e.g., JWT or OAuth).

File Uploads: Fully implement the handling of uploaded medical documents.

Expanded Disease Models: Scale the platform to support additional chronic conditions beyond diabetes.

Analytics Dashboard: Provide insights for doctors on query volumes and response times.

👥 The Team

Patrick Musyoka - Team Leader / Backend Developer

Min Thiha Tun - Backend Developer

Patricia Ramos - Graphic Designer / Frontend Developer

Eman Rashid - Presentation and Documentation

Medical AI Assistant Demo App

=======

🩺 Assist AI - Medical Assistant Platform

38e8ee4 (WIP: save local changes before pull)