Building a WhatsApp Flight Search Chatbot with Python and Twilio

The Idea

What if you could search for flights just by sending a WhatsApp message? No app downloads, no sign-ups — just chat naturally. That was the premise behind my WhatsApp Flight Chatbot.

The Tech Stack

• Flask: Lightweight Python web framework to handle incoming WhatsApp messages
• Twilio WhatsApp API: The bridge between WhatsApp and my Flask server
• AviationStack API: Real-time flight data
• Hugging Face: Natural language processing to understand user intent

How It Works

When a user sends a WhatsApp message to my Twilio number, Twilio makes a POST request to my Flask webhook with the message content.

@app.route('/webhook', methods=['POST'])
def webhook():
    message = request.form.get('Body')
    sender = request.form.get('From')
    
    response = process_message(sender, message)
    
    resp = MessagingResponse()
    resp.message(response)
    return str(resp)

The Conversation State Machine

The trickiest part of a chatbot is managing conversation state. I used a simple Redis-like in-memory dictionary to track where each user is in the conversation flow:

• State 0: Greet and ask for origin city
• State 1: Ask for destination city
• State 2: Ask for travel date
• State 3: Fetch and return flight results

user_states = {}

def process_message(sender, message):
    state = user_states.get(sender, {'step': 0, 'data': {}})
    
    if state['step'] == 0:
        state['data']['origin'] = extract_city(message)
        state['step'] = 1
        user_states[sender] = state
        return "Great! Where are you flying to?"
    # ... and so on

Natural Language Understanding with Hugging Face

Instead of requiring users to type exact city codes, I used a Hugging Face NER (Named Entity Recognition) model to extract location names from natural language. "I want to fly from Lahore to Dubai" becomes {origin: 'LHE', destination: 'DXB'}.

Key Challenges

1. Session persistence: WhatsApp messages are stateless — each message is a separate HTTP request. You need to maintain state externally.
2. Rate limits: Both Twilio and AviationStack have API rate limits. Cache results where possible.
3. User experience: WhatsApp is a text-only medium. Formatting your responses with clear structure (using WhatsApp's bold, italic markers) makes a huge difference.

Watch the demo video to see it in action.