FlavorBot
RAG Assistant
A production-grade RAG system deployed on AWS serverless architecture that transforms restaurant operations. Powered by AWS Bedrock, Lambda, and OpenSearch, this AI assistant handles customer inquiries with enterprise accuracy and 99.9% uptime.
Customer Query
Menu Assistant Active
"What are your gluten-free pasta options?"
AI Response (RAG-Powered)
We offer three gluten-free pasta dishes: Penne Arrabiata, Linguine Primavera, and Gluten-Free Carbonara. All are made with certified GF pasta. Would you like details on any specific dish?
Response Time
< 2 seconds
Accuracy Rate
97%
Customer Queries/Day
500+
Staff Time Saved
15 hrs/week
Beyond Simple FAQs
This isn't a chatbot - it's a domain-expert AI trained on your restaurant's complete knowledge base.
Menu Intelligence
Advanced RAG pipeline deployed on AWS Lambda that understands dish descriptions, ingredients, allergens, and dietary preferences. Menu PDFs stored in S3 are parsed via AWS Textract, converted to embeddings, and indexed in OpenSearch for sub-second semantic retrieval.
Reservation Assistant
24/7 serverless booking system using AWS Lambda and API Gateway. Integrates with RDS PostgreSQL for real-time availability checks, DynamoDB for session management, and EventBridge for automated confirmation emails via SES.
Customer Experience Engine
Production RAG system using AWS Bedrock (Claude) for grounded generation. ElastiCache Redis handles conversation context, CloudWatch monitors performance, and X-Ray provides distributed tracing for 99.9% uptime SLA.
Real-World Query Examples
See how the RAG system handles actual customer questions with precision and context.
Menu Recommendations
Customer Query
"What are your best vegetarian options?"
RAG System Response
The AI retrieves dishes tagged as vegetarian, ranks by popularity and chef recommendations.
Allergy Management
Customer Query
"I'm allergic to shellfish. What can I safely order?"
RAG System Response
Semantic search filters menu by allergen data, providing safe recommendations.
Reservation Intelligence
Customer Query
"Can I book a table for 8 people this Saturday at 7 PM?"
RAG System Response
Checks real-time availability database, suggests alternatives if fully booked.
Special Events
Customer Query
"Do you host private parties?"
RAG System Response
Retrieves event policy documents, pricing, capacity limits, and booking process.
RAG Pipeline Architecture
Data Ingestion Pipeline
PDFs uploaded to S3 trigger Lambda functions that use AWS Textract for OCR. Documents are chunked using LangChain, embedded via SageMaker endpoints, and indexed in OpenSearch with k-NN vector search enabled.
Semantic Search Layer
API Gateway routes customer queries to Lambda, which generates embeddings and performs k-NN search in OpenSearch. The vector database returns top-5 most semantically similar menu items with sub-100ms latency.
Context Assembly
Retrieved documents from OpenSearch are assembled with conversation history from ElastiCache Redis. The context prompt is constructed with restaurant-specific constraints and formatting instructions.
Grounded Generation
AWS Bedrock (Claude model) receives the enriched prompt via Lambda invocation. Responses are streamed back through API Gateway WebSocket, monitored by CloudWatch, and logged to S3 for compliance auditing.
Technology Stack
Production-grade infrastructure built for scale and reliability.
AI Infrastructure
- AWS Bedrock (Claude)
- LangChain
- HuggingFace Embeddings
- AWS SageMaker
Data Layer
- AWS OpenSearch (Vector Search)
- Amazon RDS PostgreSQL
- ElastiCache Redis
Frontend
- Next.js 14
- React
- Tailwind CSS
- CloudFront CDN
Cloud Services
- AWS Lambda
- API Gateway
- S3
- CloudWatch
- AWS Secrets Manager
Measurable Impact
Reduction in phone call volume
Average customer satisfaction score
Availability without staff overhead
This RAG system demonstrates how retrieval-augmented generation can transform customer service in the hospitality industry. By grounding AI responses in verified restaurant data, we eliminate hallucinations while delivering instant, accurate assistance at scale.