NovaMigrate

Agentic AI for Cloud Migration Assessment & Modernization using Amazon Nova Built for the Amazon Nova Hackathon to demonstrate how GenAI can automate early-stage cloud migration decision-making.


Inspiration

Cloud migration planning is still a manual exercise. In large programs, architects spend weeks reviewing spreadsheets, estimating complexity, and debating the right migration approach. We wanted to see if Amazon Nova could take the first pass — analyzing applications and giving architects a structured starting point instead of a blank sheet.

We asked:

Can AI act like a Cloud Architect and automate migration assessment and decision-making?

NovaMigrate brings Agentic AI intelligence into cloud transformation by using Amazon Nova to analyze applications and generate structured, enterprise-level recommendations.


What it does

NovaMigrate takes an application portfolio (Excel/CSV) and automatically:

  • Calculate Application & Database Complexity
  • Recommends the right 6R migration strategy
  • Estimates effort (T-shirt size + Effort person-weeks)
  • Provides short executive reasoning
  • Suggests modernization opportunities

Portfolio assessment time reduces from weeks to minutes, with consistent and explainable outputs.


How we built it

NovaMigrate is a Prompt-as-Policy Agent powered by Amazon Bedrock – Nova. Instead of building multiple services, we focused on a Prompt-as-Policy approach — embedding migration decision logic directly into the model.

Input Parameters

Portfolio data includes:

  • Application Architecture (Monolith / 3-Tier / Microservices)
  • App Server Count & OS Compatibility
  • Technology Stack / Middleware
  • Integration Count
  • Database Count, Type, Version, and Size

Decision Framework

We embedded enterprise-style rules directly in the prompt:

  • Complexity Scoring Rules
  • Aggregation Logic
  • 6R Strategy Mapping
  • Effort Estimation Model
  • Modernization Recommendation Logic

Amazon Nova interprets these frameworks and produces structured, deterministic outputs.


Complexity Model

Overall complexity is calculated as:

Overall Complexity = max(Application Complexity, Database Complexity)

Where:

  • Application Complexity = App-level complexity score
  • Database Complexity = DB-level complexity score

Effort Estimation

Effort is mapped based on overall complexity:

Complexity Effort (Person-Weeks)
Low 2–4
Medium 5–8
High 9–16
Very High 17+

Architecture (Current)

  • Amazon Bedrock (Nova Lite / Micro)
  • Single-shot Prompt-as-Policy execution
  • Excel/CSV portfolio input
  • Markdown-based decision knowledge base

Challenges we ran into

  • Balancing fixed migration rules with AI-based reasoning
  • Ensuring recommendations match real enterprise migration practices.
  • Generating consistent and explainable outputs

Accomplishments

  • Built a working AI-based Migration Advisor using Amazon Nova.
  • Encoded real-world cloud architect decision logic into prompts.
  • Achieved consistent results across multiple portfolio scenarios

What we learned

  • Amazon Nova performs well for structured enterprise reasoning
  • Prompt engineering is critical for accuracy and cost optimization
  • Agentic AI can automate decision workflows, not just content generation
  • Combining domain knowledge with AI improves output quality
  • Generative AI can accelerate large-scale cloud transformation

What's next – Level 2 Agentic AI

Next, we plan to evolve NovaMigrate into a multi-agent workflow where each agent handles complexity analysis, strategy selection, and validation.

  • Multi-agent decision flow
  • Confidence scoring and self-validation
  • Self-correction and confidence scoring
  • Portfolio-level insights (waves, risk clusters)

Vision: NovaMigrate evolves into a full AI-powered Cloud Transformation Co-Pilot for enterprise modernization programs.

Real-world value

NovaMigrate is designed for early-stage migration planning, where organizations need quick visibility into:

  • Overall portfolio complexity
  • Migration effort sizing
  • Strategy distribution across applications

It helps architects focus their time on high-risk systems instead of manual initial analysis.

⚠️ Notes on Demo Assumptions & Model Selection

The goal of this project is to demonstrate decision automation patterns using Amazon Nova, not to replace detailed migration discovery.

Demo Logic Disclaimer

The rules defined in our prompts for:

  • Complexity scoring
  • Migration strategy selection
  • Modernization recommendations
  • Effort estimation

are simplified for demonstration purposes only. These weights, thresholds, and multipliers are not based on real-world enterprise benchmarks.
In production environments, such models would be derived from historical migration data, enterprise standards, and detailed dependency analysis.


Model Selection: Nova Lite vs Pro / Premier

For this hackathon, we used Amazon Nova Lite to optimize cost and enable rapid experimentation.

If implemented with higher-tier models:

  • Nova Pro → Deeper reasoning and improved recommendation quality
  • Nova Premier → Stronger multi-step enterprise reasoning and higher output consistency

Nova Lite provided the right balance for a demo scenario, while Pro or Premier would further enhance enterprise-grade depth and accuracy.

Guardrails & Security Considerations

For demo simplicity, we did not implement:

  • LLM guardrails or content filtering
  • Enterprise-grade access control
  • Data encryption and secure input validation

In a production deployment, the solution would include:

  • Model guardrails and output validation
  • IAM-based access control
  • Secure data handling and logging
  • Compliance-aligned architecture design

Decision Support (Human-in-the-Loop)

NovaMigrate is designed as a decision-support system, not a fully automated migration engine. The AI-generated outputs — including migration strategy, complexity, and effort — are intended to assist:

  • Cloud Architects
  • Migration Teams
  • Enterprise Decision Makers

Final decisions should always be reviewed and validated by domain experts. In production environments:

Final Decision = AI Recommendation + Human Validation

This approach improves accuracy, reduces risk, and aligns with responsible AI practices. Security and governance would be mandatory for real enterprise use.

Repository is Private

The repository is private for security reasons. Read access has been granted to:

Both invitations are currently pending. Please let us know if there are any access issues.

Built With

Share this project:

Updates