How patrikszepesi/voice-agents works
This repository implements a hotel voice assistant and a separate Streamlit conversation-log viewer. The main workflow is guest speech captured by the audio streamer, processed through a Bedrock streaming session, and returned as spoken audio; model tool requests can perform guest-profile and reservation operations against DynamoDB. The viewer parses the assistant log into conversation, event, and token-usage displays. The excerpts establish the audio queues, tool dispatch, and database access, but do not show all tool-routing details; the tool-execution link is therefore marked unverified.
Main entry points: hotel_agent.py Bedrock Stream · db_setup.py Demo Data Setup · app.py Conversation Viewer
Components 9
Voice Assistant
- Audio I/O
hotel_agent.py - Bedrock Stream
hotel_agent.py
Hotel Operations
- Hotel Tools
hotel_agent.py - DynamoDB
- Demo Data Setup
db_setup.py
Log Viewer
- Assistant Log
- Conversation Viewer
app.py
Other components
- Guest
- Amazon Bedrock
Connections 10
- Guest → Audio I/O: speaks to
hotel_agent.py - Audio I/O → Bedrock Stream: queues audio
hotel_agent.py - Bedrock Stream → Amazon Bedrock: streams events
hotel_agent.py - Amazon Bedrock → Bedrock Stream: returns events
hotel_agent.py - Bedrock Stream → Hotel Tools: dispatches tools
hotel_agent.py - Hotel Tools → DynamoDB: reads records
hotel_agent.py - Demo Data Setup → DynamoDB: creates and seeds
db_setup.py - Bedrock Stream → Audio I/O: queues audio output
hotel_agent.py - Audio I/O → Guest: plays response
hotel_agent.py - Assistant Log → Conversation Viewer: reads and parses
app.py