This Python script sets up a bi-directional Voice AI Agent using Flask, Twilio, and OpenAI. The agent receives and processes voice input from callers in real-time, leveraging Twilio's telephony capabilities to manage the call flow and OpenAI's GPT-4 model to generate intelligent, context-aware responses. The code ensures seamless, interactive co…

voice_ai_agent.py

from flask import Flask, request, jsonify
import os
from twilio.rest import Client
from twilio.twiml.voice_response import Gather, VoiceResponse
from dotenv import load_dotenv
import openai

# Load environment variables from .env file
load_dotenv()

# Initialize Flask app
app = Flask(__name__)

# Initialize OpenAI with the API key
openai_api_key = os.environ.get('OPENAI_APIKEY')
oai = openai.OpenAI(api_key=openai_api_key)

# Twilio credentials
account_sid = os.environ.get('ACCOUNT_SID')
auth_token = os.environ.get('AUTH_TOKEN')

# Phone numbers
from_number = "+88888888"
to_number = "+1223978706"


def handle_ai_response(text):
    response = oai.chat.completions.create(
        model="gpt-4",
        messages=[
            {"role": "system", "content": "You are an assistant that responds to questions and statements."},
            {"role": "user", "content": f"Respond to the following text: {text}"}
        ]
    )
    return response.choices[0].message.content.strip()


@app.route('/handleResponse', methods=['GET', 'POST'])
def handle_response():
    caller_speech = request.values.get('SpeechResult', '')
    print(caller_speech)
    llm_completion = handle_ai_response(caller_speech)
    
    # Create a TwiML response
    resp = VoiceResponse()
    resp.say(llm_completion, voice='male')
    resp.redirect('/gather')
    return str(resp)


@app.route("/gather", methods=['GET', 'POST'])
def gather():
    response = VoiceResponse()
    gather = Gather(input='speech', action='/handleResponse', speechTimeout='auto')

    # This is the caller's voice transcript
    response.append(gather)
    response.redirect('/gather')
    return str(response)


@app.route("/answer", methods=['GET', 'POST'])
def answer():
    """Respond to incoming phone calls with a brief message."""
    # Generate a conversation using OpenAI GPT-4
    response = oai.chat.completions.create(
        model="gpt-4",
        messages=[
            {"role": "system", "content": "You are an easy-to-talk-to assistant."},
            {"role": "user", "content": "Strike up a conversation and talk to me about movies."}
        ]
    )

    generated_response = response.choices[0].message.content.strip()
    print(f"Response Generation Node: {generated_response}")

    # Create a TwiML response
    resp = VoiceResponse()
    resp.say(generated_response, voice='male')
    resp.redirect('/gather')

    # Make the call using Twilio API
    client = Client(account_sid, auth_token)
    call = client.calls.create(
        twiml=str(resp),
        to=to_number,
        from_=from_number,
        method='GET'
    )


    return jsonify({'message': 'Call initiated', 'call_sid': call.sid})


if __name__ == "__main__":
    app.run(debug=True)