A detailed look at how ShopNova uses a hybrid AWS architecture, combining deterministic low-cost NLU with Generative AI for scalable, intelligent customer service.
ShopNova uses a hybrid contact center architecture where Amazon Connect controls the voice experience, Amazon Lex V2 detects customer intent, AWS Lambda handles deterministic business logic, and Amazon Q in Connect supports generative FAQ answers and agent assistance.
In the diagram, solid arrows represent the active conversation path, while dashed arrows represent supporting lookups, analytics, or loopback behavior.
The customer calls the public ShopNova demo number. Amazon Connect answers the call, plays the IVR prompts, enables Contact Lens analytics, and sends the customer input to Amazon Lex V2 for intent detection.
When Lex detects a structured request such as order status, return initiation, or inventory availability, the request is routed to the AWS Lambda dispatcher. Lambda looks up the relevant mock commerce data and returns a clear response back into the voice flow.
The Connect flow handles control decisions such as goodbye, retry prompts, complaints, and human-agent escalation. If the customer asks for an agent or expresses frustration, the flow routes the call to BasicQueue.
For FAQ-style questions or fallback cases, Lex routes the conversation to Amazon Q in Connect. Amazon Q retrieves policy information from the S3 FAQ knowledge source and generates a natural answer for the customer.
Successful answers from Lambda, Amazon Q, or the Connect decision layer are delivered through the Voice Response step. After answering, the flow can ask whether the customer needs anything else and loop back to Lex for the next intent.
When a call is transferred to a human agent, Amazon Q Agent Assist can provide suggested responses and relevant knowledge articles inside the Agent Workspace, helping the agent resolve the issue faster.
Contact Lens captures transcripts, sentiment, and interaction categories, while CloudWatch provides logs, dashboards, and budget monitoring. These services observe the call flow without changing the customer conversation path.
Manages the IVR logic (ShopNova-MainFlow), queue management (BasicQueue), and hosts the Agent Workspace for human agents.
Configured with specific intents for e-commerce (tracking, returns, inventory) and multi-turn dialogue slot filling (ShopNovaBot).
Uses a dispatcher pattern (shopnova-lex-dispatcher) to route requests to specific Python 3.14 handlers for deterministic intents.
Uses S3 as a managed Knowledge Base to power both Q Self-Service and Q Agent Assist (recommending responses based on live transcripts).
Provides real-time speech analytics, transcription, and triggers the ShopNova_Negative_Sentiment rule when customer sentiment drops to <= -3.
S3 hosts the FAQ/Catalog assets and stores call recordings, while CloudWatch provides dashboarding (ShopNova-Monitoring), logging, and billing alerts.