AI Roadside Assistance Dispatch

BPMN Process Model — Agentic AI Workflow Overview

POC · n8n · Claude · LiteLLM · Langfuse
Process Flow Diagram
AI ROADSIDE ASSISTANCE DISPATCH SYSTEM MEMBER AI SYSTEM n8n · Claude · Mock API · Langfuse DISPATCHER No follow-up Satisfied No escalation Escalate Incident Report Response + ETA Follow-up query Full context Human reply Breakdown Occurs Report Breakdown via Web Interface 👤 Member Receive AI Response Technician + ETA 👤 Member Follow-up? Yes Ask Follow-up Coverage / Status 👤 Member Receive AI Answer Multi-turn reply 👤 Member Satisfied? No Receive Dispatcher Reply 👤 Member Issue Resolved Process Incident Intake & Logging ⚙️ n8n Analyze Incident Generate Response 🤖 Claude AI Assign Technician Calculate ETA Check Coverage Tier 🔌 Mock API Send Personalized Response 🤖 Claude AI Handle Follow-up Multi-turn Conv. Coverage Answers 🤖 Claude AI await query Escalate? Hand Off to Dispatcher Full Context ⚙️ n8n Log Interactions Audit Trail Store in DB 📊 Langfuse + PG Case Closed & Audited Escalation Received Review Case Conversation History Real-time Dashboard 👷 Dispatcher Respond Directly to Member 👷 Dispatcher Escalation Resolved 💡 ~40% of inbound calls are status inquiries → handled by AI ● Start No
Start / End Events
Member Tasks
Claude AI Tasks
n8n / Mock API Tasks
Observability Tasks
Dispatcher Tasks
Gateways (Decision)
Message Flow (cross-lane)
Sequence Flow
Architecture Components
⚙️

n8n

Visual workflow orchestrator. Handles intake processing, routes incidents, triggers sub-workflows, and manages the end-to-end automation pipeline.

Workflow Orchestration
🤖

Claude (Anthropic)

Conversational AI agent. Grounded on member data and coverage tiers. Provides accurate answers, handles multi-turn conversations, and decides when to escalate.

AI Reasoning + Conversation
🔀

LiteLLM

Model gateway providing provider flexibility. Allows switching AI models without changing application code. Manages API routing and failover.

Model Gateway
📊

Langfuse

LLM observability and tracing. Captures all AI interactions for compliance, debugging, and quality monitoring. Provides real-time dashboards.

Observability + Compliance
🔌

Mock API

Lightweight data layer simulating the real roadside provider backend. Manages member profiles, case creation, technician assignment, and ETA calculation.

Data / Case Management
🗄️

PostgreSQL

Persistent storage layer. Stores conversation histories, case records, member data, and audit logs across all system components.

Persistent Storage
Demo Test Members

Anna Müller

Premium

Full coverage — all services included. No restrictions on towing distance or assistance type.

Thomas Becker

Basic

Limited coverage — basic roadside only. Restricted towing radius, no trip interruption.

Sarah Klein

Plus

Intermediate coverage — extended towing, select trip services. Most common tier.

Key Process Facts
📞

40% Status Inquiries

40% of all inbound calls are members asking "where is my technician?" — fully handled by the AI without human intervention.

🕐

24/7 AI Coverage

The AI communication layer runs around the clock. Human dispatchers only engage for genuine exceptions and escalations.

🔒

No Hallucination

All AI responses are grounded on actual member data and coverage tier information. Claude cannot invent coverage details.

🔄

Full Context on Escalation

When a human dispatcher takes over, they receive the complete conversation history. No need to re-explain the situation.