My Experience in the AWS Nova Month Skill Sprint
This blog documents my journey through the AWS Builders Skill Sprint and the lessons I learned while building AI agents from scratch. Along the way, I explored tool calling, long-term memory, streaming, and MCP integration, while also overcoming real-world debugging and compatibility challenges.
The AWS Skill sprint tasked us with 5 challenges exploring the use of agentic ai, tool usage, and mcp and other such concepts.
Challenge 1 – Building the Foundation
The first challenge focused on creating a basic AI agent using the Strands Agents SDK with Ollama running locally. The primary objective was to understand the architecture of an agent, including model initialization, system prompts, and handling user interactions.
This exercise demonstrated that meaningful AI applications can be built without relying on cloud-hosted inference by leveraging locally hosted language models.
Key concepts explored:
Configuring a local language model
Creating an agent with the Strands SDK
Using system prompts to define agent behavior
Processing and responding to user queries
Challenge 2 – Extending the Agent with Tools
Creating an agent with the Strands SDK
Using system prompts to define agent behavior
Processing and responding to user queries
Challenge 2 – Extending the Agent with Tools
Once the basic agent was operational, the next step was to make it capable of performing real-world tasks through tool integration.
The implementation included:
A built-in calculator for accurate mathematical computations
A custom weather tool capable of retrieving live weather information
A custom age calculator that determines a person's age from their birth date
A custom weather tool capable of retrieving live weather information
A custom age calculator that determines a person's age from their birth date
Rather than hardcoding tool selection, the language model automatically decided which tool to invoke based on the user's request, demonstrating one of the core principles behind modern AI agents.
Key concepts explored:
Tool calling
Creating custom tools using decorators
External API integration
Intelligent tool selection by the language model
Challenge 3 – Introducing Persistent Memory
Creating custom tools using decorators
External API integration
Intelligent tool selection by the language model
Challenge 3 – Introducing Persistent Memory
Traditional conversational systems lose context once a session ends. This challenge addressed that limitation by integrating persistent semantic memory.
Using mem0 together with a FAISS vector store, the agent gained the ability to store and retrieve information across sessions.
This enabled the assistant to remember:
User names
Preferences
Previously shared information
Long-term conversational context
Preferences
Previously shared information
Long-term conversational context
While implementing this functionality, several dependency and compatibility issues had to be resolved, providing valuable experience in configuring AI development environments and debugging library interactions.
Key concepts explored:
Persistent memory
Vector databases
Semantic retrieval
Long-term personalization
Challenge 4 – Building a Complete AI Agent
Vector databases
Semantic retrieval
Long-term personalization
Challenge 4 – Building a Complete AI Agent
The fourth challenge combined all previously implemented features into a single integrated system.
The resulting agent was capable of:
Performing mathematical calculations
Fetching live weather information
Calculating ages
Remembering user information across sessions
Streaming responses in real time
Fetching live weather information
Calculating ages
Remembering user information across sessions
Streaming responses in real time
This challenge highlighted how multiple independent components can work together to create a significantly more capable and interactive assistant.
Additional debugging was required to accommodate changes introduced in newer versions of the Strands SDK, particularly with callback handlers and streaming functionality.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article