🤖 Agentic AI: Product Recommendation Chatbot with Amazon Bedrock Agents

A hands-on workshop project demonstrating how to build an intelligent, serverless e-commerce chatbot using Amazon Bedrock Agents. The chatbot autonomously makes decisions, calls APIs, and takes actions to help users find and purchase gifts through natural language interaction.

Target Audience: Solution Architects, Software Designers, and Developers Duration: ~1 hour Region: US West (Oregon) — us-west-2

---

📐 Architecture

The solution uses a fully serverless architecture:

Service	Role
Amazon Bedrock Agents	Orchestrates the conversational AI
AWS Lambda	Executes business logic and API calls
Amazon DynamoDB	Stores product and cart data
Amazon S3	Hosts knowledge base documents
Knowledge Bases for Amazon Bedrock	Provides RAG (Retrieval Augmented Generation) capabilities
Amazon Personalize	ML-based product recommendations (simulated)

---

🚀 Workshop Sections

1. 🛍️ Product Recommendation Agent

Components Created:

• DynamoDB table: producttableandapi-ws-Products-XXXX
  • Attributes: product_name (partition key), category, gender, occasion
  • Contains 100 sample products
• Lambda: GetProductDetailsFunction — retrieves filtered product data
• Lambda: PopulateProductsTableFunction — generates sample data

Agent Configuration:

• Name: product-recommendation-agent
• Model: Anthropic Claude Sonnet 4.5
• Action Group: get-product-recommendations (OpenAPI schema)
• User Input: Enabled — agent asks clarifying questions

Result: Agent asks about recipient, occasion, and preferences, then recommends relevant products.

---

2. 🛒 Cart Actions

New Components:

• DynamoDB table: producttableandapi-ws-Cart-XXXXX (user_id, product_name)
• Lambda: GetCartFunction — retrieves cart items
• Lambda: AddToCartFunction — adds products to cart

New Action Groups:

1. get-cart — retrieves cart for a user ID
2. add-item-to-cart — adds products to user's cart

Result: Agent can add products to cart, track user ID across conversation, and display cart contents.

---

3. 🎯 Amazon Personalize Integration

Component:

• Lambda: GetPersonalizeRecommendationFunction — simulates "Customers who bought X also bought Y" recommendations

Action Group: get-amazon-personalize-recommendation

Result: Agent proactively suggests related products with social proof to increase average cart size.

---

4. 🎁 Gift Wrapping Knowledge Base (RAG)

Components:

• S3 bucket with Gift-wrapping.txt
• Knowledge Base: GiftWrappingKnowledgeBase
• Data Source: GiftWrappingDataSource

Process:

1. Synchronize data source in Knowledge Bases console
2. Add knowledge base to agent
3. Agent queries knowledge base based on cart contents

Result: Agent suggests creative, context-aware gift wrapping ideas using Retrieval Augmented Generation (RAG).

---

5. 🤝 Multi-Agent Collaboration

Implements a supervisor-collaborator pattern with specialized agents.

Specialized Agents:

A. Product Details Agent

• Name: product-details-agent
• Focus: Product exploration and recommendations
• Action Group: get-product-recommendations
• Alias: get-product-alias

B. Cart Management Agent

• Name: cart-management-agent
• Focus: Cart operations (add/retrieve)
• Action Groups: add-item-to-cart, get-cart
• Alias: cart-agent-alias

Supervisor Agent:

• Name: shopping-supervisor-agent
• Multi-agent collaboration: ENABLED
• Role: Intelligently routes requests to the appropriate specialized agent

Collaborators:

Collaborator	Agent	When Invoked
product-recommender	product-details-agent	User needs product recommendations (called first)
cart-manager	cart-management-agent	User wants to add items or view cart

Conversation History: Enabled on both collaborators to maintain context, prevent redundant questions, and ensure smooth transitions.

---

🧠 Key Concepts Demonstrated

Concept	Description
Agentic AI	Autonomous, goal-driven AI systems
Action Groups	APIs that agents can invoke
Knowledge Bases	RAG for enterprise/domain data
Multi-Agent Collaboration	Specialized agents working together
Conversation History	Context sharing between agents
OpenAPI Schemas	Defining agent capabilities
Orchestration	Agent decision-making and reasoning

---

🛠️ Technologies Used

• Amazon Bedrock Agents
• Anthropic Claude Sonnet 4.5
• AWS Lambda
• Amazon DynamoDB
• Amazon S3
• Knowledge Bases for Amazon Bedrock
• Amazon Personalize
• OpenAPI Schemas
• AWS IAM Roles

---

📸 Screenshots

Screenshots from the AWS Console are included in the /images folder, covering:

• Model selection (Claude Sonnet 4.5)
• Agent details configuration
• Action group creation
• OpenAPI schema editor
• Test agent conversation window
• Trace/debugging interface
• DynamoDB table contents
• Multi-agent collaboration settings
• Conversation history toggles

---

📋 Summary

This project progressively builds a production-ready conversational AI system:

1. ✅ Basic product recommendations via Bedrock Agent
2. ✅ Cart management with DynamoDB
3. ✅ ML-based personalization with Amazon Personalize
4. ✅ Knowledge base integration for gift wrapping ideas (RAG)
5. ✅ Multi-agent collaboration with supervisor pattern

---

👤 Author

Ahmad Sultani

• 🔗 LinkedIn
• ☁️ AWS Partner: Generative AI Essentials (December 2025)
• 🎯 AWS Solutions Architect Associate (SAA-C03) — In Progress