Flask Voice AI Backend Build

I’m rolling out Merchants Virtual, an automated answering service that takes emergency plumbing and electrical calls in the UK, and I need the full Python/Flask infrastructure assembled on Replit. I will supply the voice script and have the Twilio number ready; your role is to connect every technical piece so a caller can interact seamlessly with Vapi.ai through Twilio and I can later review exactly what happened.
Build a Complete AI Answering Service Backend (Vapi.ai + Twilio + Replit Dashboard)
This is a "Full Stack" job (Backend + Frontend Dashboard + AI Configuration). It’s small but requires competence.
Project Overview:
I am building "Merchants Virtual," an AI answering service for UK emergency trades (plumbers/electricians). I need a developer to build the entire technical infrastructure.
I will handle the Voice Prompt/Scripting and Phone Number procurement.
You will handle everything else (Backend, Database, Dashboard, API Integrations).
Budget:
$500 (Fixed Price)
• Time to Complete: 5 - 7 Days.
o Day 1-2: Setup & Backend.
o Day 3-4: Dashboard & Twilio Integration.
o Day 5: Handover & Testing.
Project Overview:
I am building "Merchants Virtual," an AI answering service for UK emergency trades (plumbers/electricians). I need a developer to build the entire technical infrastructure.
I will handle the Voice Prompt/Scripting and Phone Number procurement.
You will handle everything else (Backend, Database, Dashboard, API Integrations).
The Stack (Non-Negotiable):
• Host: Replit (Python/Flask).
• Voice AI: Vapi.ai.
• Telephony: Twilio.
• Database: SQLite (Simple, persistent file storage).

Scope of Work (The Deliverables):
1. The Backend (Python/Flask on Replit)
• Set up a Flask server with a /webhook route to receive "End of Call" reports from Vapi.
• Logic: Parse the JSON payload to extract:
o caller_number
o transcript
o summary
o recording_url
o urgency_score (1-10)
• Routing:
o If urgency_score > 7: Send SMS via Twilio + Email via SMTP.
o If urgency_score < 7: Send Email only.
2. The Dashboard (Frontend)
• Create a clean, mobile-responsive HTML dashboard at /dashboard.
• Security: Simple Hardcoded Password Login (Environment Variable).
• Features:
o Table view of all calls.
o Status badges (Urgent / Routine).
o Audio Player: Embedded player to listen to the call recording directly from the table.
o Search: Filter by date or phone number.
3. Vapi "Tool" Configuration (JSON)
• I will write the system prompt (the script).
• You must write the "Function Calling" definitions (JSON) for Vapi to ensure it accurately extracts:
o Postcode (UK Format validation).
o Urgency_Boolean (True/False).
o Issue_Summary (Short text).
4. Handover & Documentation
• The code must use os.environ for all API Keys (Twilio SID, Auth Token, Vapi Private Key).
• Deliverable: A Replit "Join Link" or Forkable Repl.
• Video Guide: A 5-minute Loom video showing me exactly where to paste my keys to make it go live.
Acceptance Criteria:
1. I can simulate a call to the Vapi agent.
2. The dashboard updates instantly with the call details and audio.
3. If I simulate an emergency, I receive an SMS.
4. The code is clean, commented, and hosted on my Replit account.