š From a Simple AI Agent to NOVA MENTOR AI ā My Journey of Building a Knowledge-Powered Assistant with AWS
My journey of developing NOVA MENTOR AI ā a knowledge-powered Generative AI assistant using Amazon Bedrock, Amazon Nova Pro, Strands Agents SDK, memory systems, tool calling, streaming responses, and knowledge retrieval.
NOVA MENTOR AI architecture workflow built with Amazon Bedrock, Amazon Nova Pro, Strands Agents SDK, Mem0, FAISS, Tools, and Knowledge Base.
š From a Simple AI Agent to NOVA MENTOR AI ā My Journey of Building a Knowledge-Powered Assistant with AWS
A few days ago, I started my journey of exploring Generative AI, AI agents, and intelligent cloud applications through the AWS Builders Skill Sprint Challenges.
Before starting this project, I thought building an AI assistant was mainly about connecting a model, writing prompts, and generating responses.
But while developing this project, my perspective completely changed.
I realized that modern AI applications are not just chatbots.
They are intelligent systems that can understand requests, use tools, remember information, retrieve knowledge, and provide meaningful responses.
This idea became the foundation of my project:
š NOVA MENTOR AI
An end-to-end Generative AI assistant built using Amazon Bedrock, Amazon Nova Pro, Strands Agents SDK, Memory Systems, Tool Calling, Streaming Responses, and Knowledge Retrieval.
šÆ Starting the Journey
My goal for this project was simple:
Build an AI assistant step-by-step and understand how real-world AI agent architectures work behind the scenes.
I didn't want to just make the code run.
I wanted to understand how each component works together to create an intelligent system.
Through this journey, I gradually transformed a simple AI agent into a complete knowledge-powered assistant.
My final AI system is capable of:
š¤ Understanding user conversations
š ļø Performing actions using external tools
š§ Remembering user context
ā” Providing streaming responses
š Retrieving information from knowledge sources
š ļø Performing actions using external tools
š§ Remembering user context
ā” Providing streaming responses
š Retrieving information from knowledge sources
š¤ Challenge 1 ā Building My First AI Agent
The first step was creating a basic AI assistant using Strands Agents SDK.
I integrated:
ā
Amazon Bedrock
ā Amazon Nova Pro
ā Strands Agents SDK
ā Amazon Nova Pro
ā Strands Agents SDK
For the first time, I was interacting with a foundation model directly through my own implementation.
When NOVA MENTOR AI generated its first successful response, it was an exciting moment.
It was a simple conversational agent, but it became the foundation for everything I built next.
š ļø Challenge 2 ā Giving My Agent Real Capabilities
The second challenge introduced tool calling.
Instead of only generating text responses, I wanted my assistant to perform useful actions.
So I added:
š§® Calculator Tool
š¤ļø Weather Tool
š Age Calculator Tool
š¤ļø Weather Tool
š Age Calculator Tool
Now my AI agent could understand the request, choose the required tool, execute it, and provide accurate results.
This was the stage where my project started feeling less like a chatbot and more like an intelligent assistant.
š§ Challenge 3 ā Adding Memory Capabilities
The third challenge focused on making the assistant remember information.
Real assistants should maintain context instead of treating every conversation as new.
For memory implementation, I worked with:
ā
Mem0
ā FAISS Vector Store
ā FAISS Vector Store
I added persistent memory so NOVA MENTOR AI could store user preferences and deliver more personalized responses.
During this stage, I understood the importance of memory systems and vector databases in AI applications.
ā” Challenge 4 ā Creating a Smart AI Agent
Challenge 4 combined everything together.
My assistant now had:
š§ Memory
š ļø Tool Calling
ā” Streaming Responses
āļø Amazon Bedrock Integration
š ļø Tool Calling
ā” Streaming Responses
āļø Amazon Bedrock Integration
This stage connected all individual components into a single intelligent workflow.
NOVA MENTOR AI was no longer just answering questions.
It could analyze requests, use tools, remember previous interactions, and respond more naturally.
š Challenge 5 ā Building a Knowledge-Powered Assistant
The final challenge focused on knowledge retrieval.
I created a custom knowledge base and connected it with my assistant.
This allowed NOVA MENTOR AI to provide responses based on specific information sources.
The final assistant could:
š¤ Understand requests
š ļø Execute tools
š§ Maintain memory
ā” Stream responses
š Retrieve knowledge
š ļø Execute tools
š§ Maintain memory
ā” Stream responses
š Retrieve knowledge
At this stage, my project had successfully evolved from a simple AI agent into a complete Knowledge-Powered AI Assistant.
š Learning Beyond the Challenges
One of the biggest influences during this journey was the AWS User Group Madurai community.
Kodai Cloud Kondattam was more than just a meetup.
It was an environment that encouraged learning through participation.
The event included:
āļø AWS Skill Builder activities
š ļø Hands-on sessions
š Community engagement
š¤ Discussions around AI and emerging technologies
š ļø Hands-on sessions
š Community engagement
š¤ Discussions around AI and emerging technologies
A memorable highlight was seeing Bumblebee ā the Agentic AI DeepRacer, introduced as the first non-human chief guest at an AWS community meetup in India.
Experiences like these make learning memorable.
More importantly, they motivate students to take the next step and start building projects of their own.
Looking back now, attending Kodai Cloud Kondattam became the spark that started this entire journey.
āļø Technologies Explored
Throughout this project, I worked with:
ā
Amazon Bedrock
ā Amazon Nova Pro
ā Strands Agents SDK
ā Python
ā Mem0
ā FAISS
ā Tool Calling
ā Streaming Responses
ā Knowledge Retrieval Systems
ā AI Agent Architecture Design
ā Amazon Nova Pro
ā Strands Agents SDK
ā Python
ā Mem0
ā FAISS
ā Tool Calling
ā Streaming Responses
ā Knowledge Retrieval Systems
ā AI Agent Architecture Design
š” Biggest Takeaway
The biggest lesson I learned is that building AI systems is not only about writing prompts.
It is about designing complete intelligent systems.
Understanding workflows.
Connecting different components.
Managing memory.
Retrieving accurate knowledge.
Debugging problems.
Improving the system continuously.
Each challenge helped me understand how modern Generative AI applications are developed.
š Future Improvements
I am planning to improve NOVA MENTOR AI by exploring:
š¹ MCP Integration
š¹ Advanced RAG Pipelines
š¹ AWS Documentation Search
š¹ OpenSearch Integration
š¹ Multi-Agent Systems
š¹ Enterprise AI Assistants
š¹ Advanced RAG Pipelines
š¹ AWS Documentation Search
š¹ OpenSearch Integration
š¹ Multi-Agent Systems
š¹ Enterprise AI Assistants
š Acknowledgements
A special thank you to everyone who supported and inspired me throughout this journey.
š Vivek Raja P S
š Poobalan P
š Logesh S
š Madhumithaa
š Poobalan P
š Logesh S
š Madhumithaa
for their continuous motivation, encouragement, and support.
š¤ Karthikeyan Chinnadurai
š¤ Thamaraikannan G
š¤ Harini G
š¤ Thamaraikannan G
š¤ Harini G
for their valuable guidance and technical support whenever I faced challenges.
Special appreciation to AWS User Group Madurai for creating opportunities that encourage students to learn, build, experiment, and grow.
š» GitHub Repository
https://github.com/SRIGANTH-K/nova-mentor-ai
Still Learning.
Still Building.
Still Exploring. š
Still Building.
Still Exploring. š
Built with ā¤ļø using Amazon Bedrock, Amazon Nova Pro, Strands Agents SDK, and Python.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article