🏥 Patient-Chat

A chat application that leverages Langchain, Langgraph, Langsmith and Neo4j knowledge graph.

The patient (Nikola Tesla) mentions that he is taking lisinopril twice a day. But current patient profile states he is taking them once and that's a mismatch. Our intelligent agent detects it and sends a medication change request to the doctor (Shown in bottom left).

Besides, conversation summary and medical insights of the patient is shown on the right.

🌟 Product Features

• 🤖 Doctor assistant AI chat for patients
• 📝 Chat summarization to reduce long context window cost and time
• 🕸️ Knowledge Graph utilization for RAG techniques using Neo4j
• 🔧 Function calling for external systems and APIs (appointment scheduling, medication changes)
• 📊 Comprehensive chat summarization for patients
• 🩺 Medical history insights based on patient preferences, diet, and diagnostics
• 🔄 LLM agnostic design

📁 Project File Structure

📁 Patient_chat
📁 home
  📁 constants
  📁 langchains
  📁 function_tools
  📁 models
📁 notebooks
📁 static
📁 templates

🚀 Setup

Follow these instructions to set up the project locally:

🐍 Python Version

Tested on Python 3.12.6 and above. We can't guarantee compatibility with earlier versions.

🌐 Virtual Environment

1. Create a virtual environment named venv:

   python -m venv venv
   

2. Activate it:

   source venv/bin/activate
   

📦 Install Dependencies

pip install -r requirements.txt

🔑 Environment Variables

Create a .env file at the project root. See env.example for reference. Required variables:

ANTHROPIC_API_KEY=your-api-key
GOOGLE_API_KEY=your-api-key
OPENAI_API_KEY=your-api-key

LANGCHAIN_API_KEY=your-api-key
LANGCHAIN_TRACING_V2=true
LANGCHAIN_ENDPOINT=your-langchain-endpoint
LANGCHAIN_PROJECT=your-langchain-project-name

NEO4J_URI=your-neo4j-url
NEO4J_USERNAME=your-neo4j-user-name
NEO4J_PASSWORD=your-neo4j-user-password

🗄️ Database Setup - PostgreSQL

We use PostgreSQL for storing patient information and chat history. Two main tables:

1. patient: Patient bio and medical information
2. chat_history: Chat history, thread_id, and summarized chat history

Setup steps:

1. Install PostgreSQL from postgresql.org
2. Navigate to the db_scripts folder:

   cd db_scripts
   

3. Make scripts executable:

   chmod +x create_db_tables.sh insert_data.sh
   

4. Run scripts:

   ./create_db_tables.sh
   

   ./insert_data.sh
   

5. Configure database connection in settings.py:

   DATABASES = {
     'default': {
         'ENGINE': 'django.db.backends.postgresql_psycopg2',
         'NAME': 'patient_db',
         'USER': '',
         'PASSWORD': '',
         'HOST': '127.0.0.1',
         'PORT': '5432'
     }
   }

🕸️ Knowledge Graph

1. Install Neo4j or use the server version
2. Set environment variables in .env:

   NEO4J_URI=your-neo4j-url
   NEO4J_USERNAME=your-neo4j-user-name
   NEO4J_PASSWORD=your-neo4j-user-password
   

3. Note: We start with an empty knowledge graph and load data gradually with each user chat.

🚀 Run Django Server

If you are still in db_scripts folder, switch back to root project directory.

Start the server:

python manage.py runserver

The server should start running at http://127.0.0.1:8000/

🧪 Testing

Run tests:

python manage.py test

Project Architecture

Below we provide overall project architectural details.

Chain for Chat and Summary

Chain for Knowledge Graph

💬 Long Chat Optimizations

To reduce the cost and time from long chat input-output context, we are using the summarization technique. We filter and summarize chat history in the backend, storing summaries in the database using a unique thread_id in Langsmith.

🔄 Changing AI Models

Support for OpenAI, Anthropic, and Google Gemini is included. For other langchain AI libraries:

1. Add the model's langchain dependency in requirements.txt
2. Add the API Key environment variable in settings.py
3. (Optional) Add an entry in constants.py
4. Add the actual API Key to the .env file

🛠️ Function Calling

• Change appointment date
• Medication change request

🚀 Future Improvements

• 🌊 Streaming chat
• ⚡ Parallel API calls to reduce interaction time
• 📜 Display previous chat threads in the UI (already in the database)