Building the MAP Portal: How AI-Powered Development Accelerated Our AWS Migration Hub

Building the MAP Portal: How AI-Powered Development Accelerated Our AWS Migration Hub

Executive Summary

From a workshop conversation at AWS re:Invent to a production-ready application in just 14 days of active development—spanning 35 calendar days including the Christmas holidays. That’s the power of AI-driven development.

We built a comprehensive AWS Migration Acceleration Program (MAP) portal featuring 14 content pages, interactive calculators, AI-powered chatbot, user authentication, and cloud-synced personalization. From idea to working MVP: 4 days. From MVP to production: 14 days of active development.

The secret? Leveraging Kiro CLI and an AI-driven Software Development Lifecycle (AI-SDLC) that we learned about at AWS re:Invent 2025.

This is the story of how a real business problem, a workshop insight, and AI-assisted development came together to fundamentally transform our approach to software delivery.

The Genesis: From re:Invent Workshop to Real-World Solution

The Spark: AWS re:Invent 2025

During a workshop on Kiro and AI-SDLC at AWS re:Invent 2025, we had an “aha moment.” The presenters demonstrated how specialised AI agents could handle different phases of the software development lifecycle—from requirements gathering to deployment. But it was just a demo, a proof of concept.

Then came the real business problem.

The Real Business Challenge

Our sales and delivery teams were facing a critical problem: AWS MAP knowledge was siloed across the organisation, creating confusion and missed opportunities.

The Pain Points:

• Qualification Confusion: Sales teams were unclear when an opportunity qualified for MAP funding
• Inconsistent Guidance: Different teams had different interpretations of MAP requirements
• Knowledge Silos: MAP expertise was locked in the heads of a few specialists
• Business Impact: Projects were being delayed, opportunities were being lost, and revenue was at risk as clients weren’t hitting growth targets without proper MAP positioning and tagging wasn’t being applied to track progress
• Execution Uncertainty: Delivery teams were unclear on requirements for each MAP phase (Assess, Mobilise, Migrate & Modernise)
• Funding Complexity: Understanding MAP funding eligibility and calculation was a black box

The Impact:

• Opportunities not being qualified correctly for MAP funding
• Inconsistent customer conversations about migration benefits
• Delays in project execution due to unclear phase requirements
• Underutilisation of available MAP funding
• Frustrated teams asking the same questions repeatedly

What We Needed:

• A single source of truth for MAP programme information
• Clear qualification criteria for MAP funding eligibility
• Step-by-step guidance through each MAP phase
• Tools to calculate ROI and funding amounts
• Self-service access so teams could find answers without waiting for specialists
• A centralised hub accessible to sales, delivery, and customer-facing teams

We needed an interactive, comprehensive portal that would democratise MAP knowledge across the organisation.

But we had constraints:

• Timeline: Needed quickly to support active opportunities in the pipeline
• Quality: Had to be production-ready and trustworthy, not a prototype
• Adoption: Had to be intuitive enough that teams would actually use it

This was the perfect opportunity to test what we learnt at re:Invent.

The Experiment: Could AI-SDLC Really Work?

We decided to put AI-SDLC to the test with a real project, not a demo. The hypothesis: If we could get from idea to working MVP in a week, we’d know this approach was viable.

The Timeline That Changed Everything:

• Day 0 (Dec 8, 2025): Workshop at re:Invent, idea conceived
• Day 1-2: Set up Kiro CLI, configured specialised agents
• Day 3-4: Built MVP with core features (content pages, navigation, basic calculators)
• Day 4 (Dec 12, 2025): Working MVP deployed ✅
• Dec 15 - Jan 11: Down period due to varying circumstances (holidays, other commitments)
• Jan 12 - Jan 12: Final production refinements
• Day 14 of active development (Jan 12, 2026): Production-ready solution deployed ✅

Note: This timeline spanned 35 calendar days but only 14 days of active development, including a down period from Dec 15 - Jan 11 due to Christmas holidays and other circumstances.

The Requirements (evolved from MVP to production):

• 14 comprehensive content pages covering all MAP phases
• Interactive ROI and Funding calculators with PDF export
• AI chatbot powered by AWS Bedrock for instant answers
• User authentication with AWS Cognito
• Cloud-synced bookmarks and preferences
• Admin analytics dashboard
• WCAG 2.1 Level AA accessibility compliance
• Mobile-responsive design with PWA capabilities
• Production-ready infrastructure on AWS

Traditional Development Timeline: 3-4 months minimum

AI-SDLC Timeline: 4 days to MVP, 14 days of active development to production

Acceleration: 6-8x faster

The Solution: AI-SDLC with Specialized Agents

Instead of following traditional waterfall or even agile methodologies, we adopted an AI-driven Software Development Lifecycle using Kiro CLI — Amazon’s AI-powered development assistant. The key innovation? Specialised AI agents that mirrored our development team structure.

Meet the AI Team

We configured five specialised Kiro agents, each with distinct responsibilities:

1. Product Owner Agent

• Gathered requirements and defined user stories
• Created comprehensive product intent documentation
• Prioritized features based on business value
• Maintained the product vision throughout development

2. Domain Architect Agent

• Designed system architecture and component relationships
• Defined API contracts and data models
• Created technical specifications for each feature
• Ensured architectural consistency across the application

3. Technical Architect Agent

• Translated domain architecture into implementation details
• Designed AWS infrastructure using CDK
• Specified technology stack and frameworks
• Created detailed technical architecture documents

4. Full-Stack Engineer Agent

• Implemented frontend React components
• Built backend Lambda functions and APIs
• Integrated AWS services (Cognito, DynamoDB, Bedrock)
• Wrote production-ready code following best practices

5. Deployment Engineer Agent

• Orchestrated infrastructure deployment
• Configured frontend with live endpoints
• Validated end-to-end functionality
• Resolved deployment issues autonomously

The AI-SDLC Workflow

Each feature followed a systematic progression through specialised agents:

Product Owner → Domain Architect → Technical Architect → Full-Stack Engineer → Deployment Engineer

Example: AI Chatbot Feature (v2.2.0)

1. Product Owner defined the requirement: “Users need instant answers to MAP-related questions without leaving the portal”

2. Domain Architect designed the solution:

   • RAG (Retrieval-Augmented Generation) architecture
   • 8-section knowledge base covering all MAP content
   • Conversational interface with floating button
   • Cost-efficient implementation target: <$5/month

3. Technical Architect specified implementation:

   • AWS Bedrock with Claude 3 Haiku model
   • Lambda Function URL with CORS
   • In-memory knowledge base with keyword matching
   • Frontend integration with React hooks

4. Full-Stack Engineer built the feature:

   • Lambda function with Bedrock SDK integration
   • React ChatBot component with conversation state
   • Knowledge base with 8 MAP sections
   • Error handling and loading states

5. Deployment Engineer deployed and validated:

   • CDK stack deployment (MapPortalChatBotStack)
   • Frontend configuration with live endpoint
   • End-to-end testing with sample queries
   • Cost monitoring setup

Result: AI chatbot deployed in 3 days, from concept to production.

The Business Benefits: Why This Matters

1. 10x Development Velocity

The Numbers That Matter:

• Idea to MVP: 4 days (Dec 8-12, 2025)
• MVP to Production: 10 additional days of active development (after down period Dec 15 - Jan 11)
• Total Active Development: 14 days
• Calendar Time: 35 days (including down period)
• Traditional Estimate: 90-120 days
• Acceleration Factor: 6-8x faster

But speed isn’t the only metric. The quality remained high:

• Zero critical bugs in production
• 98 Lighthouse accessibility score
• 90+ performance score
• WCAG 2.1 Level AA compliant

The Holiday Factor: Traditional development would have stalled during the Christmas holiday period. With AI-SDLC, we could pause and resume development seamlessly—picking up exactly where we left off without losing context.

2. Consistent Architecture & Documentation

Every feature came with comprehensive documentation:

• Product intent documents
• Domain architecture specifications
• Technical architecture details
• Implementation plans
• Deployment guides

This documentation wasn’t an afterthought—it was generated as part of the AI-SDLC process. The Domain Architect agent created technical specs before code was written, ensuring consistency and maintainability.

3. Reduced Context Switching

Traditional development requires engineers to switch between roles:

• Writing requirements
• Designing architecture
• Implementing code
• Deploying infrastructure
• Writing documentation

With specialized AI agents, each phase was handled by an agent optimized for that task. Human developers focused on validation, refinement, and strategic decisions rather than boilerplate work.

4. Cost Efficiency

Development Cost Savings:

• Traditional approach: 3 developers × 3 months = 9 person-months
• AI-SDLC approach: 1 developer × 0.5 months (14 days including holidays) = 0.5 person-months
• Savings: 94% reduction in development time

The re:Invent ROI: The workshop investment paid for itself in the first week. What we learned enabled us to deliver a production application in the time it would normally take to complete requirements gathering.

Infrastructure Cost:

• Monthly AWS costs: $9-27/month
• AI chatbot: ~$1-5/month
• Total operational cost: <$30/month

ROI: The portal is positioned to deliver value by:

• Reducing time spent answering MAP qualification questions
• Enabling sales teams to self-qualify opportunities without specialist involvement
• Accelerating deal cycles with clear MAP funding calculations
• Improving win rates by providing consistent, accurate MAP guidance to customers

Expected Impact (as adoption grows):

• Democratised MAP knowledge across the organisation
• Faster opportunity qualification
• Consistent customer conversations
• Better utilisation of MAP funding opportunities

5. Autonomous Issue Resolution

The Deployment Engineer agent didn’t just deploy—it fixed issues autonomously:

Example Issue: Frontend build failed due to missing environment variables

Traditional Approach:

1. Developer notices build failure
2. Investigates error logs
3. Identifies missing .env file
4. Creates .env with correct values
5. Rebuilds and redeploys Time: 30-60 minutes

AI-SDLC Approach:

1. Deployment agent detects build failure
2. Analyzes error: “VITE_COGNITO_USER_POOL_ID is not defined”
3. Checks deployment plan for expected outputs
4. Creates .env file with live values from infrastructure
5. Rebuilds and continues deployment Time: 2 minutes (autonomous)

Real-World Impact: The Numbers

Development Metrics

Feature Delivery Timeline (from MVP to Production)

Production Metrics (Early Adoption - Week 1)

• Uptime: 99.9%
• Page Load Time: <2 seconds (95th percentile)
• Lighthouse Scores: 98 (accessibility), 92 (performance)
• Early Feedback: Positive reception from initial users
• Portal Status: Live and accessible to organisation
• Publicised: Late January 2026 (early adoption phase)

Technical Architecture: Built for Scale

Frontend Stack

• React 18 with Vite for blazing-fast builds
• React Router v6 for client-side routing
• AWS Amplify for Cognito integration
• Recharts for interactive visualizations
• jsPDF for PDF export functionality

Backend Services

• AWS Lambda (Node.js 18) for serverless compute
• AWS Bedrock (Claude 3 Haiku) for AI chatbot
• AWS Cognito for user authentication
• Amazon DynamoDB for user data persistence
• API Gateway for RESTful APIs

Infrastructure

• AWS CDK (Python) for infrastructure as code
• Amazon S3 for static hosting
• Amazon CloudFront for global CDN
• AWS Route 53 for DNS management
• AWS Certificate Manager for SSL/TLS

Cost Optimization

• On-demand DynamoDB: Pay only for what you use
• Lambda Function URLs: No API Gateway costs for simple endpoints
• CloudFront caching: Reduced origin requests by 85%
• S3 lifecycle policies: Automatic cleanup of old versions

Result: Production-grade infrastructure for <$30/month

The AI-SDLC Advantage: Key Learnings

1. Specialised Agents > General-Purpose AI

We initially tried using a single AI assistant for all tasks. The results were inconsistent—sometimes brilliant, sometimes confused. By creating specialised agents with focused responsibilities, we achieved:

• Consistent output quality: Each agent became an expert in its domain
• Reduced hallucinations: Narrow scope meant fewer opportunities for errors
• Better context management: Agents maintained focus on their specific phase

2. Documentation as a First-Class Artifact

In traditional development, documentation is often an afterthought. With AI-SDLC, documentation was generated as part of the process:

• Product Owner created intent documents
• Domain Architect produced technical specifications
• Deployment Engineer generated deployment guides

Result: 25+ pages of comprehensive documentation, always up-to-date.

3. Autonomous Issue Resolution

The Deployment Engineer agent could fix 80% of deployment issues without human intervention:

• Configuration errors (wrong paths, missing environment variables)
• Missing dependencies (npm install, pip install)
• Infrastructure permissions (IAM grants, CORS headers)
• Build errors from incorrect paths

Human intervention required only for:

• Application logic changes
• API contract modifications
• Architecture decisions

4. Iterative Refinement

AI agents didn’t get everything right the first time. But they excelled at iterative refinement:

• User provides feedback
• Agent analyzes the issue
• Agent proposes a fix
• User validates
• Repeat until resolved

Example: The AI chatbot initially had generic responses. After 3 iterations of refinement with the Full-Stack Engineer agent, it provided accurate, context-aware answers grounded in MAP content.

Challenges and How We Overcame Them

Challenge 1: Agent Context Management

Problem: Agents would sometimes lose context between sessions, leading to inconsistent decisions.

Solution:

• Created structured handoff documents between agents
• Each agent produced artifacts that the next agent consumed
• Maintained a central AI-DLC/ directory with all agent outputs

Challenge 2: Over-Engineering

Problem: The Technical Architect agent sometimes proposed overly complex solutions.

Solution:

• Added constraints to agent prompts: “Prefer simple deployment methods for prototypes”
• Human review of architecture before implementation
• Iterative simplification based on actual requirements

Challenge 3: Deployment Validation

Problem: Automated deployment succeeded, but end-to-end functionality failed.

Solution:

• Mandatory user testing phase before deployment completion
• Deployment Engineer agent presents deployment and waits for user confirmation
• One-by-one issue resolution until user confirms “everything works”

The Future: Where AI-SDLC is Heading

Short-Term (Next 6 Months)

• Enhanced RAG: Vector search with embeddings for more accurate chatbot responses
• Automated Testing: AI agents that write and execute test cases
• Performance Optimization: AI-driven analysis of bottlenecks and automatic fixes

Medium-Term (6-12 Months)

• Multi-Agent Collaboration: Agents working together on complex features
• Continuous Learning: Agents that learn from production metrics and user feedback
• Predictive Maintenance: AI that identifies potential issues before they occur

Long-Term (12+ Months)

• Autonomous Development: AI agents that can take a product idea from concept to production with minimal human intervention
• Self-Healing Systems: Infrastructure that detects and fixes issues automatically
• AI-Driven Product Management: Agents that analyze user behavior and suggest new features

Getting Started with AI-SDLC

Want to adopt AI-SDLC for your projects? Here’s how to start:

1. Install Kiro CLI

# Install Kiro CLI (AWS's AI development assistant)
npm install -g @aws/kiro-cli

# Initialize in your project
cd your-project
kiro-cli init

2. Create Specialised Agents

Define agents in .kiro/agents/ directory:

{
  "name": "deployment-engineer-agent",
  "description": "Deploys and validates applications",
  "prompt": "You are a Deployment Engineer...",
  "tools": ["read", "write", "shell"],
  "model": "claude-sonnet-4.5"
}

3. Establish AI-SDLC Workflow

Create a structured workflow:

1. Product Owner defines requirements
2. Domain Architect designs solution
3. Technical Architect specifies implementation
4. Full-Stack Engineer builds features
5. Deployment Engineer deploys and validates

4. Maintain Documentation Structure

Create an AI-DLC/ directory for agent artifacts:

AI-DLC/
├── {feature}/
│   ├── product-owner/
│   │   └── intent.md
│   ├── domain-architect/
│   │   └── domain-architecture.md
│   ├── technical-design/
│   │   └── technical-architecture.md
│   ├── code-generation/
│   │   ├── implementation_plan.md
│   │   └── deployment_handoff.md
│   └── deployment/
│       ├── deployment-plan.md
│       └── progress-checklist.md

5. Start Small

Don’t try to AI-ify your entire development process at once:

• Start with one feature
• Use one or two specialised agents
• Iterate and refine based on results
• Gradually expand to more agents and features

Conclusion: The New Era of Software Development

What started as a workshop conversation at AWS re:Invent 2025 became a production application in just 14 days of active development (35 calendar days including a down period). The MAP Portal project proved that AI-SDLC isn’t just theory—it’s a practical approach that delivers real business value.

The Journey:

• Day 0 (Dec 8, 2025): Workshop at re:Invent, learnt about Kiro and AI-SDLC
• Day 4 (Dec 12, 2025): Working MVP deployed
• Dec 15 - Jan 11: Down period (holidays and other circumstances)
• Day 14 of active development (Jan 12, 2026): Production-ready application serving real users

The Results:

• 6-8x faster development without sacrificing quality
• 14 days of active development (35 calendar days with down period)
• 94% reduction in development time compared to traditional approaches
• Comprehensive documentation generated as part of the process
• Autonomous issue resolution for 80% of deployment problems
• Production-ready infrastructure for <$30/month
• Seamless pause/resume - No context loss during down period

But the real value isn’t in the metrics—it’s in the transformation of the development experience. Developers spend less time on boilerplate and more time on strategic decisions. Documentation is always up-to-date. Deployments are predictable and reliable. And most importantly, ideas can become reality in days, not months.

The future of software development is here, and it’s powered by AI.

A Note on AWS re:Invent

This project wouldn’t have been possible without the insights gained at AWS re:Invent 2025. The workshop on Kiro and AI-SDLC provided the foundation, but more importantly, it changed how we think about software development.

Key Takeaway: Conferences aren’t just about learning—they’re about sparking ideas that solve real business problems. The ROI of attending re:Invent was realised in less than a week.

Attending re:Invent 2026? Look for workshops on AI-assisted development, Kiro CLI, and AI-SDLC methodologies. The insights you gain could transform your next project.