Simple POC using LangChain to built a WhatsApp bot chat with.

Requirements

• Twilio Account
  • Go to Twilio and sign up for a free account.
  • Activate the Twilio Sandbox for WhatsApp
  • Join the Sandbox:
    • Twilio provides a unique WhatsApp number and a code to connect.
    • Add the Twilio-provided WhatsApp number to your WhatsApp contacts.
    • Send the unique code (e.g., join sandbox-code) to the Twilio number from WhatsApp to join the Sandbox.
  • Configure Webhooks for Incoming and Outgoing Messages, URL where Twilio will send incoming messages. We will use the ngrok server endpoint for local testing.
• ngrok
  • We will use ngrok for local testing
• OpenAI account
  • We will use this for the LLM, you can replace it with Ollama if needed.
  • Once u have the account, get the API key and copy the value in the .env.example.
  • Now update the .env.example to .env

Local Development

• Install the dependencies by running pip install -r requirement.txt
• Run the following to start the local ngrok server ngrok http http://localhost:8080
• The port 8080 is mapped from your local python server. Copy the url shown in the Forwarding in the Twilio console for Whatsapp Webhooks for Incoming and Outgoing Messages.
• Now start the python server python server.py

User flow

• User can interact with the LLM via bot to ask generic questions.
• User can upload pdf to the bot and ask questions around it.

Code

• The python server exposes an endpoint which Twilio is consuming.
• We are exposing one endpoint for both chat and file upload.

@app.route('/whatsapp/incoming', methods=['POST'])
def whatsapp():
    incoming_msg = request.values.get('Body', '').strip()  # User's WhatsApp message
    incoming_file_url = request.values.get('MediaUrl0')  # File URL if provided

    response = MessagingResponse()
    msg = response.message()

    # Temp config, for prod can ue the incoming number
    config = {"configurable": {"thread_id": "abc123"}}

    try:
        if incoming_file_url:
            file_name: str = handle_file_download(incoming_file_url)
            file_content = extract_text_from_file(file_name)

            # Index the extracted content
            index_document(file_content)

            msg.body("File content indexed successfully. Now you can ask questions related to the uploaded document.")
        else:
            # Use the LangChain agent to handle the user's message
            input_messages = [HumanMessage(incoming_msg)]
            print(f"input_messages: {input_messages}")

            agent_response = agent.invoke({"messages": input_messages}, config)
            ai_response = agent_response["messages"][-1]
            msg.body(str(ai_response.content))
    except Exception as e:
        msg.body(f"Oops! Something went wrong: {e}")

    return str(response)

if __name__ == '__main__':
    # starting the server at 8080
    app.run(port=8080, debug=True)

UI