This virtual assistant platform is a comprehensive real-time speech-to-speech
conversational AI system built on AWS and LiveKit with Amazon Bedrock Nova Sonic
for voice and Amazon Bedrock AgentCore Runtime for chat. While designed as a
hotel guest services assistant, this platform is easily adaptable to any
industry by simply changing system prompts and integrating industry-specific
tools. It provides natural, context-aware conversations through multiple
interfaces including voice, chat, and web applications.
A real-time conversational AI platform built on AWS with Amazon Bedrock Nova Sonic for voice and Amazon Bedrock AgentCore Runtime for chat. While designed as a hotel guest services assistant, this platform is easily adaptable to any industry by simply changing system prompts and integrating industry-specific tools via MCP servers.
2026-10-28
ดาว
18
เครื่องมือ
—
อัปไทม์
—
Key Adaptability Features:
Industry-Agnostic Architecture: Core platform works for hospitality,
retail, healthcare, finance, and more
Simple Customization: Adapt to any industry by modifying system prompts
and available tools
Flexible Tool Integration: Easy integration with any backend system via
standardized MCP protocol
Reference Implementation: Complete hotel use case demonstrates platform
capabilities
Key Value Propositions
Multi-Modal Customer Interactions: Support voice calls (LiveKit), web
chat, and text messaging across any industry
Real-Time Natural Speech Processing: Natural, conversational interactions
with Amazon Nova Sonic
Easy Industry Adaptation: Customize for any vertical by changing system
prompts and tools - no code changes required
Backend Service Integration: Easy integration with any backend system via
standardized MCP protocol
Scalable AWS Architecture: Auto-scaling infrastructure with Amazon ECS with AWS Fargate
and Aurora Serverless
Complete Reference Implementation: Hotel use case demonstrates full
platform capabilities
Room Management: Check availability, make reservations, modify bookings
Guest Services: Concierge information, amenities details, service requests
Housekeeping Coordination: Schedule services, track requests, room status
updates
Multi-Language Support: English and Spanish with localized system prompts
Conversation Memory: Persistent context across interactions using
Amazon Bedrock AgentCore Memory
Note: These hotel-specific capabilities demonstrate the platform's flexibility.
The same architecture supports any industry by replacing the hotel PMS tools
with industry-specific integrations.
You are responsible for the cost of the AWS services used while running this
solution. As of December 2025, the cost for running this prototype with the
default settings in the US East (N. Virginia) Region is approximately $342.01
per month for 500 voice calls and 500 chat conversations.
The costs shown represent pricing without consideration of AWS free tier
benefits.
We recommend creating a budget through
AWS Cost Explorer
to help manage costs. Prices are subject to change. For full details, refer to
the pricing webpage for each AWS service used in this solution.
The following table provides a sample cost breakdown for deploying this solution
with the default parameters in the US East (N. Virginia) Region for one month:
Cost estimate (500 voice calls, 500 chat conversations per month)
Component Overview
Component
Description
Cost [USD]
Backend Integration & Database
Industry-specific tools + knowledge base
$6.78
Virtual Assistant Voice (LiveKit)
AI voice processing infrastructure
$292.89
Virtual Assistant Chat
AI chat processing + simulated messaging
$42.34
Total
$342.01
Backend Integration & Database - $6.78/month
AWS Service
Usage
Cost [USD]
Amazon Cognito
User authentication
$3.00
AWS Secrets Manager
API key management
$1.70
Amazon CloudWatch
Monitoring & logs
$1.00
AWS Lambda
Serverless compute
$0.84
Bedrock AgentCore Gateway
Tool calling gateway
$0.10
S3 Standard
Object storage & vector store
$0.06
Amazon Bedrock - Amazon Titan Text Embeddings V2
Text embeddings for RAG
$0.04
Amazon DynamoDB
Hotel PMS data storage (6 tables)
$0.04
Total
$6.78
Virtual Assistant Voice (LiveKit) - $284.63/month
Note: Voice infrastructure is optional. If you only need chat/messaging, you
can skip LiveKit configuration and save $284.63/month (including $81.30 in VPC
costs).
AWS Service
Usage
Cost [USD]
Amazon Bedrock - Nova Sonic 2
AI voice processing (500 calls)
$65.80
Network Address Translation (NAT) Gateway
Network address translation (VPC)
$74.00
LiveKit Cloud
Real-time voice infrastructure (500 calls)
$70.00
AWS Fargate
Containerized compute
$60.00
Public IPv4 Address
Public IP addresses (VPC)
$7.30
Amazon CloudWatch
Monitoring & logs
$5.00
Data Transfer
Network data transfer
$2.00
AWS Secrets Manager
API key management
$0.53
Total
$284.63
Virtual Assistant Chat - $28.34/month
AWS Service
Usage
Cost [USD]
Amazon Bedrock - Amazon Nova 2 Lite (global inference)
AI chat processing (500 conversations)
$14.80
Amazon API Gateway
Simulated messaging API
$0.02
Amazon Cognito
User authentication
$2.00
DynamoDB
Message storage & message buffering
$0.27
Bedrock AgentCore Memory
Conversation memory storage
$2.50
Amazon CloudWatch
Monitoring & logs
$1.50
Bedrock AgentCore Runtime
Agent execution runtime
$0.76
Bedrock AgentCore Observability
Agent performance monitoring
$0.35
AWS Step Functions
Message buffering workflow orchestration
$0.71
AWS Lambda
Message processing & buffering handlers
$0.02
AWS Secrets Manager
API credentials
$0.40
AWS WAF Web ACL
API protection
$5.00
S3 Standard
Configuration storage
$0.01
Total
$28.34
Alternative Messaging Configuration:
For production WhatsApp messaging via AWS End User Messaging Social (instead of
simulated messaging backend):
AWS Service
Usage
Cost [USD]
AWS End User Messaging Social
WhatsApp message delivery (5,000 msgs)
~$50.00
AWS Lambda
Message processing
$0.04
Amazon CloudWatch
Monitoring & logs
$0.50
Additional Cost
~$50.54
Note: WhatsApp Business API pricing varies by region and message type. This
estimate assumes average costs.
The actual cost will vary based on:
Number of voice calls and chat conversations processed
Average duration of voice calls (default estimate assumes 10 minutes)
Complexity of conversations and token usage
Selected AWS region for deployment
Whether production backend integration replaces simulated database
Important: This estimate excludes telephony and messaging integration costs.
Voice telephony (Twilio/Amazon Connect) and messaging connectors (WhatsApp/SMS)
require separate providers. When using the optional WhatsApp integration with
AWS End User Messaging Social, additional costs apply based on message volume
and WhatsApp Business API pricing.
Deployment and Development
Project Structure
This project is organized as an NX monorepo with pnpm workspaces, providing
efficient build caching, task orchestration, and dependency management across
multiple technologies:
The Virtual Assistant Platform is designed for easy adaptation to any industry.
The hotel implementation serves as a complete reference.
What you customize:
System prompts and conversation flows
Industry-specific tools via MCP protocol
Knowledge base content
Backend API integrations
What stays the same:
Core infrastructure (voice, chat, messaging)
AWS architecture and deployment
Authentication and security
The platform uses the Model Context Protocol (MCP) to connect your existing APIs
as tools. AgentCore Gateway wraps any REST API with an OpenAPI specification,
eliminating the need to implement MCP protocol yourself.
# Install pnpm
curl -fsSL https://get.pnpm.io/install.sh | sh -
# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh
# Install required tools and versions
pnpm env use --global 22
uv python install 3.14
# Install global packages
npm install -g aws-cdk
Activate Models in Amazon Bedrock
This prototype uses Amazon Nova and Anthropic Claude models on Amazon Bedrock.
Amazon models are enabled by default, while Anthropic models require a
marketplace subscription.
Recommended approach: Subscribe to all Anthropic models to ensure
compatibility with current and future model versions:
# Subscribe to all Anthropic models (recommended)
uv run scripts/bedrock_subscribe.py "Anthropic*" \
--company-name "Your Company Name" \
--use-cases "AI-powered virtual assistant and conversational AI"
# Subscribe to specific Anthropic model only
uv run scripts/bedrock_subscribe.py "Anthropic Claude 3.5 Haiku"
The prototype requires these models:
Amazon Nova 2 Sonic (enabled by default)
Amazon Nova 2 Lite (enabled by default)
Amazon Titan Text Embeddings V2 (enabled by default)
Configure LiveKit Project Secret
Create LiveKit Project:
Navigate to project settings:
Create a new API key:
Copy the generated API key and secret:
Store the LiveKit credentials in AWS Secrets Manager with the secret name
"virtual-assistant-livekit":
# Replace with your actual values from the LiveKit dashboard
aws secretsmanager create-secret \
--name "virtual-assistant-livekit" \
--description "LiveKit credentials for Virtual Assistant" \
--secret-string '{
"LIVEKIT_URL": "wss://your-project-name-xxxxxxxx.livekit.cloud",
"LIVEKIT_API_KEY": "your-api-key-from-step-4",
"LIVEKIT_API_SECRET": "your-api-secret-from-step-4"
}'
Configure WhatsApp Integration (Optional)
The Virtual Assistant Platform supports real WhatsApp messaging through AWS End
User Messaging Social (EUM Social). This is optional - the system works with
simulated messaging by default.
Prerequisites for WhatsApp Integration
AWS End User Messaging Social set up in your account
nx deploy infra --context eumSocialTopicArn=arn:aws:sns:us-east-1:123456789012:whatsapp-messages --context eumSocialPhoneNumberId=phone-number-id-01234567890123456789012345678901
For detailed WhatsApp integration setup, cross-account configuration, and
troubleshooting, see the WhatsApp Integration Guide
(documentation/whatsapp-integration.md).
Build and Deploy
The entire build and deployment can be done with just a few commands:
pnpm install
pnpm exec nx bootstrap infra # First time only
pnpm exec nx deploy infra # Deploy to AWS (~2.5 min)
The deploy command will automatically trigger the necessary builds and then
deploy the infrastructure.
CDK Context Variables
The infrastructure supports various configuration options through CDK context
variables. These can be set in packages/infra/cdk.context.json or passed via
command line.
Core Configuration
bedrock_xacct_role (Optional): Cross-account IAM role ARN for Bedrock
access when models are in a different AWS account
Find your User Pool using the VirtualAssistantStack.UserPoolId from the
stack deployment outputs.
In the "Users" section, click "Create user" and follow the instructions.
Use the Chat Assistant
Navigate to http://localhost:5173 after running nx serve demo.
Conversation suggestions
Look for a hotel
You: "¿Dónde puedo hacer meditación maya con cuencos tibetanos?" Agent: should
respond with information about Paraíso Tulum You: "¿Qué hay de rico para comer
allí?" Agent: should respond with information about Paraíso Tulum
Book a Room
You: "Quiero reservar una habitación para dos personas." Agent: should collect
required details and respond with a booking confirmation.
Request Maintenance
You: "¿Puedes hacer mantenimiento en la habitación 101?"
Check Out
You: "¿Puedes hacer checkout de la habitación 101?"
Cleanup
In the event that you decide to stop using the prototype, we recommend that you
follow a tear-down procedure. Most of the services used have no cost when there
is no active use with the notable exception of storage in S3 and CloudWatch
Logs. AWS CloudFormation via CDK will tear down all resources except for Amazon
S3 buckets and AWS CloudWatch Logs log groups with data.
1. Destroy CDK Stacks
nx run infra:destroy
2. Clean Up Remaining S3 Buckets
Due to versioning, lifecycle rules are the most thorough approach for complete
cleanup.
# Find virtual-assistant related bucket names
ASSISTANT_BUCKETS=$(aws s3api list-buckets --query 'Buckets[?contains(Name, `virtualassistant`) || contains(Name, `cdk-hnb659fds`)].Name' --output text)
# Apply lifecycle rules to expire all objects and versions after 1 day
for bucket in $ASSISTANT_BUCKETS; do
echo "Setting lifecycle rule for bucket: $bucket"
aws s3api put-bucket-lifecycle-configuration \
--bucket "$bucket" \
--lifecycle-configuration '{
"Rules": [{
"ID": "DeleteEverything",
"Status": "Enabled",
"Filter": {},
"Expiration": {"Days": 1},
"NoncurrentVersionExpiration": {"NoncurrentDays": 1},
"AbortIncompleteMultipartUpload": {"DaysAfterInitiation": 1}
}]
}'
done
# After 24+ hours, delete the empty buckets
for bucket in $ASSISTANT_BUCKETS; do
echo "Deleting bucket: $bucket"
aws s3 rb s3://"$bucket"
done
3. Delete CloudWatch Log Groups
# Delete virtual-assistant related log groups
aws logs describe-log-groups \
--query 'logGroups[?contains(logGroupName, `VirtualAssistant`) || contains(logGroupName, `HotelPms`)].logGroupName' \
--output text | xargs -I {} aws logs delete-log-group --log-group-name {}
Important Notes:
The CDKToolkit stack should remain as it's used by other CDK deployments
Wait 24+ hours after setting lifecycle rules before attempting to delete S3
buckets
Troubleshooting
For common issues and debugging guidance, see the
troubleshooting guide. This includes
specific guidance for WhatsApp integration, voice processing, deployment issues,
and performance optimization.
Improvements
As part of the prototype process, we have some
improvement ideas for your path to
production.
Amazon Connect Integration: If your use case requires telephone support,
human agent escalation, or complete contact center capabilities, we recommend
using Amazon Connect. The
AWS Workshop for Amazon Connect Integration
provides comprehensive guidance for building virtual assistants with Amazon
Connect, including integration with Amazon Bedrock Nova Sonic for natural
conversations.
Sample code, software libraries, command line tools, proofs of concept,
templates, or other related technology are provided as AWS Content or
Third-Party Content under the AWS Customer Agreement, or the relevant written
agreement between you and AWS (whichever applies). You should not use this AWS
Content or Third-Party Content in your production accounts, or on production or
other critical data. You are responsible for testing, securing, and optimizing
the AWS Content or Third-Party Content, such as sample code, as appropriate for
production grade use based on your specific quality control practices and
standards. Deploying AWS Content or Third-Party Content may incur AWS charges
for creating or using AWS chargeable resources, such as running Amazon EC2
instances or using Amazon S3 storage.
S
Sample Voice And Chat Hotel Assistant
aws-samples·18 ดาว·
A real-time conversational AI platform built on AWS with Amazon Bedrock Nova Sonic for voice and Amazon Bedrock AgentCore Runtime for chat. While designed as a hotel guest services assistant, this pla
A real-time conversational AI platform built on AWS with Amazon Bedrock Nova Sonic for voice and Amazon Bedrock AgentCore Runtime for chat. While designed as a hotel guest services assistant, this platform is easily adaptable to any industry by simply changing system prompts and integrating industry-specific tools via MCP servers.
Sample Voice And Chat Hotel Assistant ต้องใช้ API key ไหม?
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
Sample Voice And Chat Hotel Assistant ใช้ทรานสปอร์ตแบบใด?
Sample Voice And Chat Hotel Assistant ทำงานแบบโลคัลผ่าน stdio
Sample Voice And Chat Hotel Assistant ถูกตรวจสอบล่าสุดเมื่อใด?
ตรวจสอบล่าสุดเมื่อ 2026-10-28 A real-time conversational AI platform built on AWS with Amazon Bedrock Nova Sonic for voice and Amazon Bedrock AgentCore Runtime for chat. While designed as a hotel guest services assistant, this platform is easily adaptable to any industry by simply changing system prompts and integrating industry-specific tools via MCP servers.