# Project export: PETSOS This document was generated by HackStack to give an AI agent context about a hackathon project. Sections are labeled with their provenance; content marked as truncated was cut to keep this document small. ## Project metadata - Hackathon: UC Berkeley AI Hackathon 2025 - Tagline: Instantly connect panicked pet owners to life-saving guidance and nearby emergency vetsβ€”just by speaking. - Devpost: https://devpost.com/software/petsos-0iqmaz - GitHub: https://github.com/ashleyvarghesee/petsos - Demo: https://www.canva.com/design/DAGrGmql6rI/JA0bI8kjSJowPQ0TGuKOjw/edit?utm_content=DAGrGmql6rI&utm_campaign=designshare&utm_medium=link2&utm_source=sharebutton - Team: 1 GitHub contributor(s) β€” ashleyvarghese (1 commits) ## Devpost submission (written by the team) ### Inspiration This project was inspired by my family friend's loss of their dog during a critical moment of crisis. It shed light on an overlooked issue: emergency pet health service. As a pet owner, it terrifies me that there is no 911 for our loved pets. So with the power of AI - I took on the challenge. ### What it does PET911 is a voice-activated emergency assistant that guides pet owners through life-threatening situations β€” completely hands-free. It simulates a real 911-style call for pets and responds with calm, intelligent, and timely support. Here's what it does: πŸ—£οΈ Understands your voice in real-time Users can call a real phone number and speak naturally. PET911 uses voice streaming through Vapi to transcribe speech instantly and maintain a fluid, multi-turn conversation. πŸ€– Provides life-saving guidance using AI Powered by Claude, the assistant triages emergencies by asking crucial questions and delivering context-specific, step-by-step instructions (e.g., performing the Heimlich maneuver for a choking dog based on its breed and weight). 🧠 Remembers what you said The assistant uses memory from the conversation so it doesn’t repeat itself or forget details like the pet’s condition or size β€” enabling natural, human-like dialogue even in high-stress moments. 🌐 Finds the closest emergency vet clinic PET911 uses your ZIP code or live coordinates to identify and link you to the closest open emergency clinicβ€”ensuring fast, location-aware care when every second counts. πŸ“ Sends an SMS with directions PET911 identifies a nearby emergency clinic and sends a text message with a direct link to the location. This allows the user to begin navigation immediately β€” skipping the stress of searching online or fumbling with maps during a crisis. πŸ”„ Transfers to a live vet If needed, PET911 offers to transfer the call to a pre-set backup assistant, simulating the experience of speaking with a real vet receptionist in the area β€” for needed escalation. ❀️ Offers emotional support The AI is designed to be calm, empathetic, and reassuring β€” because pet emergencies are overwhelming, and users need to feel safe and heard. ### How we built it PET911 runs on a modular multi-agent system built with a Flask backend. A voice interface powered by Vapi captures real-time input, which is routed through specialized agents: a Triage Agent for rapid emergency assessment, a Memory Agent for contextual continuity, and a Decision Agent to determine next steps. Claude serves as the core reasoning engine. For location support, we integrated OpenStreetMap, and emergency routing is handled via SMS through Textbelt. ### Challenges we ran into We hit roadblocks integrating Fetch.ai’s Python SDK due to package conflicts, which led us to rethink how we handled real-time vet discovery. Vapi's real-time voice streaming also came with unexpected 404 webhook issues and delayed responses, requiring deep debugging. On top of that, prompting Claude for step-by-step, non-overwhelming instructions demanded careful prompt engineering to simulate a calm, conversational emergency assistant. ### Accomplishments we're proud of Built a fully functional voice AI emergency system from scratch using Vapi and Claude, with natural, multi-turn memory and contextual awareness. Integrated a simulated agent ecosystem combining triage, decision-making, and location support under emergency pressure. Engineered real-time responses that adapt to user input instead of overwhelming themβ€”just like a real 911 operator would. Successfully routed emergency info via SMS, with geolocation-based vet directions and calming follow-ups. Debugged complex streaming issues and webhooks under time pressure and got everything running seamlessly across tools. ### What we learned Building PET911 taught me how to design for urgency, empathy, and clarity under pressure. I deepened my skills in real-time voice interaction, multi-agent system design, and external API integration. I also learned how to troubleshoot complex streaming and webhook errors, all while keeping the user experience as the top priorityβ€”especially in life-or-death scenarios. ## README (from the GitHub repository) # 🐾 PET911 - AI-Powered Pet Emergency Assistant A voice-activated AI assistant for pet emergencies with real-time SMS notifications and vet locator services. ## πŸš€ Features ### Core Functionality - **Voice-Activated Emergency Response** - Natural conversation with pet owners during emergencies - **Real-Time SMS Notifications** - Automatic vet locator links sent via SMS when calls start - **Intelligent Triage System** - Step-by-step first-aid guidance for common pet emergencies - **Live Vet Transfer** - Seamless transfer to emergency veterinarians when needed - **Bilingual Support** - English and Spanish language support - **Vet Locator Service** - Web-based emergency clinic finder with directions ### Technical Features - **Real-Time Decision Logic** - Dynamic webhook processing with intelligent event routing - **Robust Phone Number Detection** - Recursive payload analysis for reliable SMS delivery - **Quota Management** - Smart SMS limiting to prevent API abuse - **Error Handling** - Graceful fallbacks and comprehensive logging - **Memory-Free Design** - Consistent, reliable responses every time ## πŸ› οΈ Tech Stack - **Backend**: Flask (Python) - **Voice Platform**: Vapi.ai - **LLM**: Anthropic Claude 3 Haiku - **SMS Service**: Textbelt API - **Geolocation**: OpenStreetMap Nominatim API - **Voice Synthesis**: ElevenLabs - **Speech Recognition**: Deepgram Nova-2 ## πŸ“‹ Prerequisites - Python 3.8+ - ngrok (for webhook tunneling) - Vapi.ai account - Anthropic API key - Textbelt API key - ElevenLabs API key ## πŸš€ Quick Start ### 1. Clone the Repository ```bash git clone https://github.com/yourusername/pet911.git cd pet911 ``` ### 2. Install Dependencies ```bash pip install -r requirements.txt ``` ### 3. Environment Setup Create a `.env` file in the root directory: ```env ANTHROPIC_API_KEY=your_anthropic_key_here TEXTBELT_API_KEY=your_textbelt_key_here ELEVENLABS_API_KEY=your_elevenlabs_key_here ``` ### 4. Start ngrok Tunnel ```bash ngrok http 5000 ``` ### 5. Configure Vapi Assistant 1. Go to [Vapi.ai Dashboard](https://dashboard.vapi.ai) 2. Create a new assistant 3. Set the webhook URL to your ngrok URL + `/vapi-webhook` 4. Configure voice settings (ElevenLabs recommended) 5. Set up phone number routing ### 6. Run the Application ```bash python app.py ``` ## πŸ“ Project Structure ``` petSOS/ β”œβ”€β”€ agents/ β”‚ β”œβ”€β”€ __init__.py β”‚ β”œβ”€β”€ llm_agent.py # Claude LLM integration β”‚ └── triage_agent.py # Emergency triage logic β”œβ”€β”€ vet_locator/ β”‚ β”œβ”€β”€ main_agent.py # Vet locator service β”‚ β”œβ”€β”€ sms_service.py # SMS integration β”‚ └── templates/ β”‚ └── index.html # Vet locator web interface β”œβ”€β”€ app.py # Main Flask application β”œβ”€β”€ requirements.txt # Python dependencies β”œβ”€β”€ test_sms.py # SMS testing utility └── README.md # This file ``` ## πŸ”§ Configuration ### SMS Settings - **Target Numbers**: Configure in `app.py` line 264 - **SMS Quota**: Adjust `MAX_SMS_PER_SESSION` in `app.py` line 30 - **Vet Locator URL**: Update `VET_LOCATOR_URL` in `app.py` line 28 ### Emergency Response - **Transfer Number**: Set in Vapi assistant configuration - **First Aid Protocols**: Customize in `agents/llm_agent.py` - **Language Support**: Configure in system prompts ## πŸ§ͺ Testing ### Test SMS Functionality ```bash python test_sms.py ``` ### Test Vet Locator 1. Start the Flask server 2. Visit `http://localhost:5000/vet-locator` 3. Enter a location to find nearby clinics ### Test Voice Assistant 1. Call your configured Vapi phone number 2. Describe a pet emergency 3. Verify SMS is sent automatically 4. Test vet transfer functionality ## πŸ“Š API Endpoints - `POST /vapi-webhook/chat/completions` - LLM chat completions - `POST /vapi-webhook` - Vapi event webhooks - `GET /vet-locator` - Vet locator web interface - `POST /vet-locator/api/clinics` - Clinic search API - `GET /vet-locator/api/health` - Health check ## πŸ”’ Security Considerations - API keys stored in environment variables - SMS quota limiting to prevent abuse - Input validation on all endpoints - Error handling without sensitive data exposure ## 🀝 Contributing 1. Fork the repository 2. Create a feature branch (`git checkout -b feature/amazing-feature`) 3. Commit your changes (`git commit -m 'Add amazing feature'`) 4. Push to the branch (`git push origin feature/amazing-feature`) 5. Open a Pull Request ## πŸ“ License This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details. ## πŸ™ Acknowledgments - **Vapi.ai** for voice platform infrastructure - **Anthropic** for Claude LLM capabilities - **OpenStreetMap** for geolocation services - **Textbelt** for SMS delivery - **ElevenLabs** for voice synthesis ## πŸ“ž Support For support, email support@pet911.com or create an issue in this repository. --- **Built with ❀️ for pet safety and emergency response** ## Detected evidence (automated analysis) Indexed codebase: 10 recognized source files, 53 KB. - Anthropic (technology) β€” detected in the code - Flask (technology) β€” detected in the code - HTML (language) β€” detected in the code - Python (language) β€” detected in the code - JavaScript (language) β€” claimed on Devpost, not found in the code ## Codebase structure (from repository index) ### Files (15 of 15) ``` .gitignore agents/__init__.py agents/llm_agent.py agents/triage_agent.py app.py env.example LICENSE README.md requirements.txt test_sms.py vet_locator/main_agent.py vet_locator/README.md vet_locator/requirements.txt vet_locator/sms_service.py vet_locator/templates/index.html ``` ### Dependencies - requirements.txt: anthropic, Flask@==3.0.0, python-dotenv@==1.0.0, requests@==2.31.0 - vet_locator/requirements.txt: flask@==3.0.0, python-dotenv@==1.0.0, requests@==2.31.0, twilio@==9.6.3 ### Recent commits (newest first) - Initial commit: PET911 AI-powered pet emergency assistant ## Key source files (fetched from GitHub, selected and truncated for size) ### requirements.txt ``` Flask==3.0.0 requests==2.31.0 python-dotenv==1.0.0 anthropic ``` ### vet_locator/requirements.txt ``` flask==3.0.0 requests==2.31.0 python-dotenv==1.0.0 twilio==9.6.3 ``` ### app.py ```python import json import time import random import string import requests import os from flask import Flask, request, Response, send_from_directory from dotenv import load_dotenv # Import the new LLM agent from agents.llm_agent import get_llm_response # Import vet locator SMS service try: from vet_locator.sms_service import sms_service VET_LOCATOR_AVAILABLE = True print("βœ… Vet locator SMS service loaded") except ImportError: VET_LOCATOR_AVAILABLE = False print("⚠️ Vet locator SMS service not available") # Load environment variables load_dotenv() app = Flask(__name__) # Textbelt SMS configuration TEXTBELT_API_KEY = os.getenv('TEXTBELT_API_KEY', 'textbelt') # Use env var or fallback to free testing key VET_LOCATOR_URL = "https://f93a-2600-1012-b225-308f-61e3-f4c0-8bcb-c74b.ngrok-free.app/vet-locator" # Track which call IDs have already triggered an SMS (in-memory, resets on server restart) sms_sent_call_ids = set() MAX_SMS_PER_SESSION = 1 # Safety limit to prevent using up all quota - only 1 SMS per session def send_textbelt_sms(phone_number, message): """ Send SMS using Textbelt API """ # Safety check - don't send if we've already sent too many if len(sms_sent_call_ids) >= MAX_SMS_PER_SESSION: print(f"⚠️ SMS quota limit reached ({MAX_SMS_PER_SESSION}). Not sending SMS to {phone_number}") return False try: # Format phone number (remove + if present) formatted_phone = phone_number.replace('+', '') if phone_number.startswith('+') else phone_number # Prepare the SMS data sms_data = { 'phone': formatted_phone, 'message': message, 'key': TEXTBELT_API_KEY } # Send SMS via Textbelt response = requests.post('https://textbelt.com/text', data=sms_data) result = response.json() if result.get('success'): print(f"βœ… Textbelt SMS sent successfully to {phone_number}") print(f"πŸ“Š Quota remaining: {result.get('quotaRemaining', 'Unknown')}") return True else: print(f"❌ Textbelt SMS failed: {result.get('error', 'Unknown error')}") return False except Exception as e: print(f"❌ Error sending Textbelt SMS: {e}") return False def send_vet_locator_sms(phone_number): """ Send vet locator link via Textbelt SMS (obfuscated with [dot] and [slash]) """ message = ( "🐾 PET911 Emergency Vet Locator\n\n" "Find nearby emergency vet clinics at: http[colon][slash][slash]f93a-2600-1012-b225-308f-61e3-f4c0-8bcb-c74b[dot]ngrok-free[dot]app[slash]vet-locator\n\n" "Stay safe with your pet! πŸ•πŸ±" ) return send_textbelt_sms(phone_number, message) def detect_poor_transcription(text): """ Detect if Vapi's transcription is likely poor quality. Returns True if the transcription seems garbled or unclear. """ if not text: return True # Check for very short responses that might be incomplete if len(text.strip()) < 3: return True # Check for repeated characters (common in poor transcription) for char in text: if text.count(char) > len(text) * 0.4: # If any character is >40% of the text return True # Check for common transcription artifacts artifacts = ['...', 'um', 'uh', 'ah', 'er', 'mm', 'hmm'] if any(artifact in text.lower() for artifact in artifacts): return True return False def send_vet_locator_link(phone_number, location=None, pet_info=None): """ Send vet locator link via SMS to caller Args: phone_number (str): Caller's phone number location (str): Optional ZIP code or coordinates pet_info (dict): Optional pet information """ if not VET_LOCATOR_AVAILABLE: print("❌ Vet locator SMS service not available") return False try: if pet_info: success = sms_service.send_emergency_vet_sms(phone_number, pet_info) else: success = sms_service.send_vet_locator_sms(phone_number, location) if success: print(f"βœ… Vet locator SMS sent to {phone_number}") else: print(f"❌ Failed to send vet locator SMS to {phone_number}") return success except Exception as e: print(f"❌ Error sending vet locator SMS: {e}") return False # MAIN LLM ENDPOINT - Now supports streaming @app.route('/vapi-webhook/chat/completions', methods=['POST']) def vapi_chat_completions(): data = request.json print("🟒 Received /chat/completions request") conversation_history = data.get("messages", []) phone_number = None if 'customer' in data and 'number' in data['customer']: phone_number = data['customer']['number'] print(f"πŸ“± User phone number: {phone_number}") # Remove memory-based greeting: do NOT use memory to personalize greeting # Only use standard greeting every time # Check the last user message for poor transcription if conversation_history: last_user_message = None for msg in reversed(conversation_history): if msg.get("role") == "user": last_user_message = msg.get("content", "") break if last_user_message and detect_poor_transcription(last_user_message): print(f"⚠️ Detected poor transcription: '{last_user_message}'") conversation_history.append({ "role": "system", "content": "The user's last message may have been poorly transcribed by the speech recognition system. If their response seems unclear or incomplete, ask them to repeat or speak more clearly." }) try: print(f"🟒 Calling get_llm_response with {len(conversation_history)} messages") llm_stream = get_llm_response(conversation_history, None) # R [truncated β€” 14398 more characters] ``` ### test_sms.py ```python import requests import json import os from dotenv import load_dotenv # Load environment variables load_dotenv() def test_textbelt_sms(): """Test Textbelt SMS functionality""" # Your phone number phone_number = "2093629161" # Remove the + for Textbelt # Test message message = "🐾 PET911 Test SMS\n\nThis is a test message from PET911 Emergency Vet Locator.\n\nFind nearby emergency vet clinics at: http[colon][slash][slash]f93a-2600-1012-b225-308f-61e3-f4c0-8bcb-c74b[dot]ngrok-free[dot]app[slash]vet-locator\n\nStay safe with your pet! πŸ•πŸ±" # Get API key from environment variable api_key = os.getenv('TEXTBELT_API_KEY', 'textbelt') # Textbelt API data sms_data = { 'phone': phone_number, 'message': message, 'key': api_key } print(f"πŸ“€ Testing SMS to: {phone_number}") print(f"πŸ“ Message: {message[:50]}...") print(f"πŸ”‘ Using key: {api_key}") print("\nπŸš€ Sending SMS...") try: # Send SMS via Textbelt response = requests.post('https://textbelt.com/text', data=sms_data) result = response.json() print(f"\nπŸ“Š Response Status: {response.status_code}") print(f"πŸ“Š Response: {json.dumps(result, indent=2)}") if result.get('success'): print(f"\nβœ… SMS sent successfully!") print(f"πŸ“Š Quota remaining: {result.get('quotaRemaining', 'Unknown')}") print(f"πŸ†” Text ID: {result.get('textId', 'N/A')}") return True else: print(f"\n❌ SMS failed: {result.get('error', 'Unknown error')}") return False except Exception as e: print(f"\n❌ Error sending SMS: {e}") return False if __name__ == "__main__": print("πŸ§ͺ PET911 SMS Test") print("=" * 50) success = test_textbelt_sms() if success: print("\nπŸŽ‰ Test completed successfully! Check your phone for the SMS.") else: print("\nπŸ’₯ Test failed. Check the error message above.") ``` ### agents/__init__.py ```python # PET911 Agents Package ``` ### agents/triage_agent.py ```python def handle_triage(user_input): """ Basic triage agent for PET911 emergency responses. This is a temporary mock implementation that will be replaced with AI logic. """ user_input_lower = user_input.lower() # Emergency keywords detection if "choking" in user_input_lower: return "What breed and size is your dog?" elif "swallowed" in user_input_lower: return "What did your pet swallow and when? This is important for determining the urgency." elif "bleeding" in user_input_lower: return "Where is the bleeding located? Is it severe or minor?" elif "poison" in user_input_lower or "toxic" in user_input_lower: return "What substance did your pet ingest? Do you have the packaging?" elif "seizure" in user_input_lower: return "How long has the seizure been going on? Is this the first time?" elif "hit by car" in user_input_lower or "accident" in user_input_lower: return "Is your pet conscious? Can they move all their limbs?" else: return "Please describe the emergency in detail so I can provide the best guidance." ``` ### vet_locator/main_agent.py ```python from flask import Flask, request, jsonify, send_from_directory import requests import os from dotenv import load_dotenv # Load environment variables load_dotenv() app = Flask(__name__) @app.route('/vet-locator') def index(): return send_from_directory('templates', 'index.html') @app.route('/vet-locator/api/clinics', methods=['POST']) def get_clinics(): """Get nearby emergency vet clinics based on location""" data = request.json location = data.get('zip') or data.get('location') if not location: return jsonify({'error': 'ZIP or location required'}), 400 try: # Use OpenStreetMap Nominatim API to find emergency vet clinics search_query = f"emergency+vet+clinic+{location}" resp = requests.get( f'https://nominatim.openstreetmap.org/search', params={ 'q': search_query, 'format': 'json', 'limit': 5, 'addressdetails': 1 }, headers={'User-Agent': 'PET911-VetLocator/1.0'} ) if resp.status_code != 200: return jsonify({'error': 'Failed to fetch clinic data'}), 500 data = resp.json() # Process and format clinic data clinics = [] for clinic in data[:5]: # Top 5 results display_name = clinic.get('display_name', 'Unknown Clinic') # Clean up the display name for better readability name_parts = display_name.split(',') clinic_name = name_parts[0] if name_parts else 'Emergency Vet Clinic' clinics.append({ 'name': clinic_name, 'address': display_name, 'lat': clinic.get('lat'), 'lon': clinic.get('lon'), 'distance': clinic.get('distance', 'Unknown') }) return jsonify({ 'clinics': clinics, 'location': location, 'count': len(clinics) }) except Exception as e: print(f"Error fetching clinics: {e}") return jsonify({'error': 'Failed to fetch clinic data'}), 500 @app.route('/vet-locator/api/health') def health_check(): """Health check endpoint""" return jsonify({'status': 'healthy', 'service': 'vet-locator'}) if __name__ == '__main__': port = int(os.getenv('VET_LOCATOR_PORT', 5000)) # Changed to 5000 to match main app app.run(host='0.0.0.0', port=port, debug=True) ``` ### vet_locator/sms_service.py ```python import os from twilio.rest import Client from dotenv import load_dotenv # Load environment variables load_dotenv() class VetLocatorSMS: def __init__(self): """Initialize Twilio client for SMS service""" self.account_sid = os.getenv('TWILIO_ACCOUNT_SID') self.auth_token = os.getenv('TWILIO_AUTH_TOKEN') self.from_number = os.getenv('TWILIO_FROM_NUMBER') # Your Vapi phone number self.vet_locator_url = os.getenv('VET_LOCATOR_URL') # Your ngrok URL if not all([self.account_sid, self.auth_token, self.from_number]): print("⚠️ Warning: Twilio credentials not found. SMS service will be disabled.") self.client = None else: self.client = Client(self.account_sid, self.auth_token) def send_vet_locator_sms(self, to_phone, location=None): """ Send SMS with vet locator link to caller Args: to_phone (str): Phone number to send SMS to location (str): Optional ZIP code or coordinates to pre-fill """ if not self.client: print("❌ SMS service not available - Twilio credentials missing") return False try: # Build the URL with optional location parameter url = self.vet_locator_url if location: url += f"?zip={location}" # Create the SMS message message_body = ( f"🚨 PET911 Emergency Vet Locator\n\n" f"Find nearby emergency veterinary clinics:\n" f"{url}\n\n" f"Click the link to find the closest emergency vet to your location. " f"Stay calm and get your pet help quickly!" ) # Send the SMS message = self.client.messages.create( body=message_body, from_=self.from_number, to=to_phone ) print(f"βœ… SMS sent successfully to {to_phone}") print(f"πŸ“± Message SID: {message.sid}") return True except Exception as e: print(f"❌ Failed to send SMS to {to_phone}: {e}") return False def send_emergency_vet_sms(self, to_phone, pet_info=None): """ Send emergency vet SMS with additional context Args: to_phone (str): Phone number to send SMS to pet_info (dict): Optional pet information for context """ if not self.client: print("❌ SMS service not available - Twilio credentials missing") return False try: # Build the URL url = self.vet_locator_url # Create personalized message if pet_info: pet_name = pet_info.get('name', 'your pet') pet_type = pet_info.get('type', 'pet') message_body = ( f"🚨 PET911 Emergency Alert\n\n" f"Emergency vet locator for {pet_name} ({pet_type}):\n" f"{url}\n\n" f"Find the nearest emergency veterinary clinic. " f"Drive safely and call ahead if possible." ) else: message_body = ( f"🚨 PET911 Emergency Vet Locator\n\n" f"Find nearby emergency veterinary clinics:\n" f"{url}\n\n" f"Get your pet help quickly. Drive safely!" ) # Send the SMS message = self.client.messages.create( body=message_body, from_=self.from_number, to=to_phone ) print(f"βœ… Emergency vet SMS sent to {to_phone}") print(f"πŸ“± Message SID: {message.sid}") return True except Exception as e: print(f"❌ Failed to send emergency vet SMS to {to_phone}: {e}") return False # Global instance for easy access sms_service = VetLocatorSMS() ``` ### agents/llm_agent.py ```python import os import json import anthropic from dotenv import load_dotenv # Load environment variables from .env file load_dotenv() # Initialize the Anthropic client try: # Note: Ensure you have ANTHROPIC_API_KEY in your .env file api_key = os.environ.get("ANTHROPIC_API_KEY") if not api_key: raise ValueError("ANTHROPIC_API_KEY not found in .env file.") client = anthropic.Anthropic(api_key=api_key) except ValueError as e: print(f"Error initializing Anthropic client: {e}") client = None # System prompt to define the AI's role and behavior SYSTEM_PROMPT = """ You are PET911, a caring emergency assistant for pet owners. Talk like a real person - warm, calm, and helpful. TRANSFER RULES: - USE the transferCall tool ONLY when someone EXPLICITLY asks to speak to a vet - Examples: "Can I speak to a vet?", "I need to talk to a vet", "Transfer me to a vet", "I want to speak to a veterinarian" - DO NOT transfer for: * Medical emergency descriptions like "my dog is choking" * Medical questions like "What do I do when she's coughing?" * Follow-up care questions like "How do I help her?" * General pet health questions - When someone asks to speak to a vet, use the transferCall tool and say nothing at all - DO NOT say "transferring you" or "let me transfer you" or anything about transferring - DO NOT make up stories about medical emergencies when someone asks to speak to a vet - DO NOT use names unless they tell you their name TALK LIKE A HUMAN: - Be conversational and warm - Don't say "here are the steps" or "first, second, third" - Don't count out loud or be robotic - Say things like "I want you to..." or "Can you do that for me?" - Check in with them: "How's that going?" or "Are you okay?" - Be encouraging: "Good job" or "You're doing great" - Keep responses SHORT - 1-2 sentences max - Don't give long speeches or pep talks - NEVER give multiple instructions in one response - Don't repeat information they already know - For thrusts: just say "Okay, give one thrust now" - no counting EMERGENCY RESPONSE: When someone describes a pet emergency, gather essential information naturally: 1. If they haven't mentioned pet type/breed/size, ask briefly 2. Ask about age if not mentioned 3. Get details about what happened and symptoms 4. Then provide step-by-step help STEP-BY-STEP GUIDANCE: - Give ONE small action at a time - Wait for their response before the next step - Be encouraging and supportive - Don't overwhelm with multiple instructions - Keep it simple and direct - For Heimlich: guide them through ONE thrust at a time, not multiple thrusts - Don't repeat pet details they already told you - Just say "Okay, give one thrust now" - no counting needed LANGUAGE: Match the user's language (English/Spanish) EXAMPLES: User: "My dog is choking" You: "I can help with that. What breed and size is your dog?" User: "Golden retriever, about 50 pounds" You: "How old is your dog?" User: "3 years old" You: "What exactly is happening? What symptoms are you seeing?" User: "He's coughing and pawing at his mouth" You: "I'm here to help. I want you to gently open your dog's mouth and look inside. Can you do that for me?" User: "I don't see anything" You: "That's okay. Now place one hand on his chest, just behind the front legs. Can you do that?" User: "Yes, I'm doing that" You: "Good. Now with your other hand, make a fist and place it just below the rib cage. Are you ready?" User: "Yes" You: "Perfect. Okay, give one thrust now." User: "I did it" You: "Great job. How's he doing? Is he coughing or breathing better?" User: "She's starting to cough up" You: "Good. Keep encouraging her to cough. Let me know if she's still struggling." User: "What do I do when she's coughing?" You: "That's actually a good sign - it means the object is coming up. Keep encouraging her to cough and let her work it out naturally. Just stay calm and supportive." BE NATURAL - talk like you're really there helping them, not like a robot giving instructions. Keep responses short and helpful. NEVER give multiple steps or long speeches. """ # Define the tool for call transfers, matching Vapi's expected format TRANSFER_TOOL = { "name": "transferCall", "description": "Transfer the call to a veterinarian", "input_schema": { "type": "object", "properties": {}, "required": [] } } def get_llm_response(messages, phone_number=None): """ Get response from Claude (no memory system) """ # Check if client is properly initialized if not client: raise ValueError("Anthropic client not initialized. Please check your ANTHROPIC_API_KEY.") # Prepare messages for Claude (filter out system messages) claude_messages = [] for msg in messages: if msg.get("role") == "user": claude_messages.append({"role": "user", "content": msg.get("content", "")}) elif msg.get("role") == "assistant": claude_messages.append({"role": "assistant", "content": msg.get("content", "")}) # Get response from Claude try: response = client.messages.create( model="claude-3-haiku-20240307", max_tokens=1000, system=SYSTEM_PROMPT, messages=claude_messages, tools=[TRANSFER_TOOL], stream=True ) return response except Exception as e: print(f"Error getting LLM response: {e}") raise ``` ### vet_locator/templates/index.html ```html Emergency Vets Near You - PET911

🚨 Emergency Vets Near You

PET911 - Find the nearest emergency veterinary clinic

Find Emergency Vet Clinics



Nearby Emergency Vet Clinics