How dcampuzano101/support-agent-demo works
This repository implements an airline customer-support voice agent with a companion FastAPI booking/refund service. The main workflow is caller speech through LiveKit, conversation triage, clarification or specialist dispatch, then a spoken reply or human handoff. Booking and refund specialists use the HTTP client to reach the backend; policy questions use semantic FAQ search. Source excerpts establish the agent routing, client operations, and backend persistence wiring. Other integrations are shown only where the excerpts support them.
Main entry points: support_graph.py Turn router · agent.py Voice session · main.py Booking and refund API · order_api.py Booking HTTP client
Components 19
Voice and triage
- Voice session
agent/agent.py - Turn router
agent/support_graph.py - Conversation triage
agent/classifier.py - Clarification response
agent/support_graph.py - Human handoff
agent/agent.py
Support specialists
- Booking lookup
agent/subagents.py - Refund handling
agent/subagents.py - Policy answers
agent/subagents.py - FAQ semantic search
agent/policy_faq.py - Booking HTTP client
agent/order_api.py
Booking backend
- Booking and refund API
mock_backend/main.py - API schemas
mock_backend/schemas.py - Booking and refund models
mock_backend/models.py - Database sessions
mock_backend/db.py
Other components
- Caller
- Relational database
- Jev classifier service
- OpenAI embeddings
- LiveKit voice runtime
Connections 20
- Caller → Voice session: speaks to
agent/agent.py - Voice session → Turn router: routes turns
agent/agent.py - Turn router → Conversation triage: triages conversation
agent/support_graph.py - Conversation triage → Jev classifier service: classifies with
agent/classifier.py - Turn router → Clarification response: asks clarification
agent/support_graph.py - Turn router → Booking lookup: dispatches booking
agent/support_graph.py - Turn router → Refund handling: dispatches refund
agent/support_graph.py - Turn router → Policy answers: dispatches policy
agent/support_graph.py - Voice session → Human handoff: hands off
agent/agent.py - Booking lookup → Booking HTTP client: requests booking
agent/subagents.py - Refund handling → Booking HTTP client: requests refund
agent/subagents.py - Policy answers → FAQ semantic search: searches FAQ
agent/subagents.py - FAQ semantic search → OpenAI embeddings: creates embeddings
agent/policy_faq.py - Booking HTTP client → Booking and refund API: calls REST routes
agent/order_api.py - Booking and refund API → API schemas: validates responses
mock_backend/main.py - Booking and refund API → Booking and refund models: queries and writes
mock_backend/main.py - Booking and refund API → Database sessions: uses sessions
mock_backend/main.py - Booking and refund models → Database sessions: uses base
mock_backend/models.py - Database sessions → Relational database: connects to
mock_backend/db.py - Voice session → LiveKit voice runtime: runs on
agent/agent.py