Building AI Agents with Amazon Nova Pro and Strands SDK: My AWS Nova Sprint Journey
This article documents my learning journey through the AWS Nova Sprint – April Edition organized by AWS User Group Madurai. It covers how I built AI-powered applications using Amazon Nova Pro on Amazon Bedrock, explored the Strands SDK, integrated tools and memory, and implemented Model Context Protocol (MCP) to create an AWS Documentation Assistant. The article shares practical code examples, key learnings, challenges encountered, and best practices for developers getting started with AI agents on AWS.
Building AI Agents with Amazon Nova Pro and Strands SDK: My AWS Nova Sprint Journey
Artificial Intelligence is rapidly changing how modern applications are built, and AWS is making it easier for developers to create intelligent, production-ready AI agents through Amazon Bedrock, Amazon Nova, and the Strands SDK. To strengthen my understanding of these technologies, I participated in the AWS Nova Sprint – April Edition, organized by the AWS User Group Madurai.
This hands-on workshop was designed as a progressive learning experience where each challenge built upon the previous one. Instead of simply learning theory, participants developed working AI applications while exploring the practical use of Amazon Nova, Amazon Bedrock, Strands SDK, and the Model Context Protocol (MCP).
Workshop Setup
Before beginning the challenges, I configured a complete development environment that included:
- Creating and configuring an AWS account
- Creating an IAM user with Bedrock permissions
- Installing and configuring the AWS CLI
- Enabling access to Amazon Nova models through Amazon Bedrock
- Setting up Python, Visual Studio Code, and Git
- Installing the Strands SDK and required dependencies
- Verifying AWS credentials and the local development environment
This initial setup provided a solid foundation for building AI applications throughout the workshop.
Challenge 1 – Building the First AI Agent
The first challenge introduced the fundamentals of AI agents using the Strands SDK with a locally hosted large language model through Ollama.
The objective was to understand the basic architecture of an AI agent, including:
- Creating an agent instance
- Defining a system prompt
- Sending prompts to the model
- Receiving generated responses
Although simple, this challenge helped establish how Strands abstracts interactions with language models while providing a clean developer experience.
Challenge 2 – Equipping the Agent with Tools
The next step was enabling the agent to perform actions rather than simply generating text.
Using Strands tools, I integrated:
- A calculator for mathematical operations
- A weather tool for simulated weather information
- An age calculator that computed age from a given birth date
This demonstrated one of the key strengths of AI agents—using external tools whenever a user request requires computation or structured logic instead of relying solely on language generation.
Challenge 3 – Introducing Persistent Memory
A conversational AI becomes significantly more useful when it can remember previous interactions.
In this challenge, I implemented a lightweight persistent memory system using JSON storage.
The agent was able to:
- Remember the user's name
- Store personal preferences
- Retrieve remembered information in future conversations
This highlighted the importance of memory in creating personalized AI experiences and demonstrated how even a simple persistence layer can greatly improve user interactions.
Challenge 4 – Building a Complete AI Agent
After learning about tools and memory separately, the fourth challenge combined these capabilities into a unified AI assistant.
The completed agent could:
- Perform mathematical calculations
- Provide weather information
- Calculate age
- Handle general conversations
- Select and invoke the appropriate tool automatically
This challenge illustrated how multiple capabilities can work together within a single intelligent application, providing a more natural and flexible conversational experience.
Challenge 5 – Building an MCP-Powered AWS Learning Assistant
The final challenge introduced one of the most exciting concepts in modern AI development—the Model Context Protocol (MCP).
Using the AWS Documentation MCP Server, I built an AI assistant capable of retrieving and using official AWS documentation during conversations.
This allowed the assistant to answer questions such as:
- Explaining AWS services in simple terms
- Providing AWS learning roadmaps
- Retrieving accurate AWS documentation
- Assisting with cloud-related concepts
Rather than depending only on the model's internal knowledge, the assistant could access external documentation dynamically, making its responses more relevant and reliable.
Key Technologies Explored
Throughout the workshop, I gained practical experience with several AWS and AI technologies, including:
- Amazon Bedrock
- Amazon Nova Pro
- Strands SDK
- Model Context Protocol (MCP)
- AWS Documentation MCP Server
- Python
- AWS CLI
- IAM
- Git and GitHub
Each technology contributed to a different aspect of building modern AI applications—from secure authentication and model invocation to tool integration and knowledge retrieval.
Challenges and Learning Experience
Like most hands-on workshops, the journey included troubleshooting and problem-solving.
Some of the practical challenges I encountered included:
- Configuring IAM users and AWS credentials
- Understanding Bedrock regions and model availability
- Resolving invalid model identifier errors
- Configuring Amazon Nova access correctly
- Integrating MCP servers with Strands SDK
- Managing Python virtual environments and project dependencies
Working through these issues provided valuable insight into real-world AI development and strengthened my understanding of the AWS ecosystem.
Final Thoughts
The AWS Nova Sprint was much more than a coding workshop—it was an end-to-end learning experience that demonstrated how modern AI agents are designed and built.
From creating a simple conversational assistant to developing an intelligent AWS Learning Assistant capable of accessing external documentation, every challenge introduced an important building block of agentic AI.
Completing all five challenges has significantly improved my understanding of Amazon Bedrock, Amazon Nova Pro, Strands SDK, tool calling, persistent memory, and the Model Context Protocol. More importantly, it has given me the confidence to continue building practical AI-powered applications on AWS.
I would like to thank the AWS User Group Madurai for organizing such a well-structured, hands-on learning experience. The progressive challenge-based approach made it easy to understand complex AI concepts while encouraging experimentation and practical implementation.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article