AI Voice Agent Tool for Logistics

A comprehensive web application that enables non-technical administrators to configure, test, and review calls made by adaptive AI voice agents for logistics operations.

🚀 Features

• Agent Configuration UI: Create and manage AI voice agent configurations with custom prompts and logic
• Call Triggering: Initiate test calls with driver information and load numbers
• Call Analysis: Review structured data extraction and full transcripts from completed calls
• Two Logistics Scenarios:
  • Driver Check-in: Status updates for loads in transit
  • Emergency Protocol: Handle emergency situations with immediate escalation

🏗 Architecture

Frontend (React + TypeScript)

• Dashboard: Single-page application with tabbed interface
• Agent Configuration: Dynamic prompt editor with scenario selection
• Call Management: Real-time call triggering and status monitoring
• Results Analysis: Structured data display with transcript review

Backend (Supabase Edge Functions)

• trigger-call: Initiates Retell AI phone calls with dynamic prompts
• retell-webhook: Handles real-time conversation and call logic and post-call processing
• Database: Supabase for agent configurations and call logs

AI Integration

• Retell AI: Voice calling platform with human-like conversation
• OpenAI GPT-4: Conversation logic and structured data extraction
• Dynamic Prompting: Context-aware agent behavior based on scenarios

📊 Database Schema

agent_configurations

• id, name, system_prompt, scenario_type, retell_settings
• Stores reusable agent configurations with optimized voice settings

call_logs

• id, driver_name, driver_phone, load_number, call_status
• full_transcript, structured_data, call_duration
• Complete call history with extracted insights

🎯 Logistics Scenarios

Scenario 1: Driver Check-in

Context: Routine status update calls to drivers about specific loads Goal: Gather current status, location, and ETA information Structured Data Extracted:

• call_outcome: "In-Transit Update" or "Arrival Confirmation"
• driver_status: "Driving", "Delayed", or "Arrived"
• current_location: Geographic location string
• eta: Estimated time of arrival

Scenario 2: Emergency Protocol

Context: Emergency situations during routine calls (breakdowns, accidents) Goal: Immediately gather critical information and escalate to human dispatcher Structured Data Extracted:

• call_outcome: "Emergency Detected"
• emergency_type: "Accident", "Breakdown", "Medical", or "Other"
• emergency_location: Precise location of emergency
• escalation_status: "Escalation Flagged"

🔧 Advanced Features

Dynamic Response Handling

• Uncooperative Drivers: Intelligent probing with graceful call termination
• Noisy Environments: Automatic retry logic with speech clarification
• Emergency Detection: Keyword-triggered protocol switching

Voice Optimization (Retell AI Settings)

• Backchanneling: Natural "mm-hmm" responses during driver speech
• Filler Words: Human-like hesitations and natural speech patterns
• Interruption Sensitivity: Balanced conversation flow management
• Response Delay: Optimized timing for natural conversation rhythm

🚦 Getting Started

Prerequisites

• Node.js 18+ and npm
• Supabase account and project
• Retell AI account with phone number
• OpenAI API key

Environment Setup

The application requires these API keys (configured via Supabase secrets):

• VITE_SUPABASE_PROJECT_ID: Supabase project ID
• VITE_SUPABASE_PUBLISHABLE_KEY: Supabase publishable key
• VITE_SUPABASE_URL: Supabase project URL (frontend use)
• SUPABASE_URL: Supabase project URL (same as above URL but for backend use)
• SUPABASE_SERVICE_ROLE_KEY: Supabase service role key (server-side)
• RETELL_API_KEY: Retell AI platform access
• RETELL_AGENT_ID: Your Retell AI agent ID

Installation

# Clone the repository
git clone <repository-url>

# Install dependencies
npm install

# Start development server
npm run dev

Supabase Setup

The application uses Supabase for:

• Database (agent configurations and call logs)
• Edge Functions (call triggering and webhook handling)
• Secrets management (API keys)

💼 Design Choices

Technology Stack Rationale

• React + TypeScript: Type-safe frontend development with excellent component reusability
• Supabase: Integrated backend-as-a-service with built-in real-time capabilities
• Retell AI: Specialized voice AI platform with superior conversation quality
• OpenAI GPT-4: Advanced natural language understanding for dynamic responses

Architecture Decisions

• Single Page Application: Single page easy to use administrator workflow
• Edge Functions: Serverless architecture for webhook handling
• Structured Data Extraction: Post-call processing for actionable insights
• Configuration-Driven: Reusable agent templates for different scenarios

Voice Quality Optimization

• Human-like Settings: Carefully tuned Retell AI parameters for natural conversations
• Context Injection: Dynamic prompt modification based on call context
• Error Recovery: Robust handling of speech recognition failures
• Emergency Protocols: Immediate conversation flow switching for urgent situations

🎬 Demo Workflow

1. Configure Agent: Create a "Driver Check-in" configuration with custom prompts
2. Trigger Call: Enter driver details (Name: "Abdullah Amer", Phone: "+923004934903", Load: "7891-B")
3. Monitor Progress: Watch real-time call status updates
4. Review Results: Analyze extracted structured data and full transcript

---