Proposal
APT Magic
A Serverless AI Platform for Personalized Image Generation and Social Interaction
1. Executive Summary
APT Magic is a serverless AI-powered web application designed to enable users to generate, personalize, and share artistic content such as AI-generated images. The platform integrates with AI foundation models via Amazon Bedrock and provides a seamless web experience using Next.js (SSR) hosted on AWS Amplify.
The MVP version focuses on real-time image generation and sharing, while the Future Design aims to scale with Bedrock agentCore/SageMaker Inference, SQS/SNS, Secret Manager & CloudTrail and AWS MLOps pipelines for advanced model orchestration and automation.
APT Magic is currently developed as a modern, cost-efficient, and secure AWS-native architecture for small to medium user bases, with planned expansion into enterprise-grade AI orchestration.
2. Problem Statement
What’s the Problem?
Most AI image generation platforms are costly, rely on opaque third-party APIs, and offer limited personalization.
Developers and creators often face high latency, lack of transparent model management, and limited control over user data security.
The Solution
APT Magic leverages AWS serverless architecture to deliver:
- Real-time AI image generation through Amazon Bedrock Stability AI models.
- Secure user authentication and content management using Amazon Cognito and DynamoDB.
- Scalable API handling via AWS Lambda and API Gateway.
- Low-latency global delivery with CloudFront CDN and WAF protection.
Future upgrades will include SQS/SNS decoupling, Bedrock AgentCore/SageMaker Inference pipelines, and cost-efficient CI/CD via CloudFormation. transforming APT Magic into a fully automated MLOps platform.
3. Solution Architecture
MVP Architecture
The MVP is a fully serverless architecture, focusing on scalability, maintainability, and cost-effectiveness.
Core AWS Services:
- Route53 + CloudFront + WAF — Secure global access and caching.
- Amplify (Next.js SSR) — Hosts the frontend and server-side rendering layer.
- API Gateway + Lambda Functions — Manage backend logic (image processing, subscription, post APIs).
- Amazon Cognito — User authentication and access control.
- Amazon S3 + DynamoDB — Data persistence and image storage.
- Amazon Bedrock — Integrates foundation model (Stability AI) for image generation.
- CloudWatch — Logging, and monitoring.
Security
- WAF + IAM policies for traffic filtering and role-based access control.
Future Design (Enhanced Architecture)
In the next phase, APT Magic will evolve into an AI orchestration platform, introducing new layers for automation, resilience, and model lifecycle management.
New Services to be Added:
Amazon SQS — For reliable message queuing between async Lambda tasks.
Amazon SNS — For real-time event notifications to users or administrators.
Amazon ElastiCache (Redis) — For rate limiting and caching of frequent inference requests.
Amazon Bedrock AgentCore — For hosting custom fine-tuned models and managing model endpoints.
CI/CD
- CloudFormation for infrastructure deployment and automation.
4. Technical Implementation
Implementation Phases
Phase 1 – MVP Deployment (Completed / Current)
- Implement Amplify (Next.js SSR) + API Gateway + Lambda.
- Integrate Bedrock Stability AI API.
- Deploy CI/CD via Gitlab CI/CD.
- Enable user authentication (Cognito) and storage (S3 + DynamoDB).
- Log and Monitor via Cloudwatch
Phase 2 – Future Design Expansion
- Introduce SQS/SNS to decouple.
- Add ElastiCache for request throttling and caching.
- Integrate Bedrock Agent to enhance AI Pipelines
- Connect GitLab Runner with CodeBuild for unified CI/CD.
5. Timeline & Milestones
| Phase | Description | Estimated Duration | Deployment Milestone |
|---|---|---|---|
| Month 1: Setup & Core API | Deploy infrastructure (IaC), Cognito, API Gateway, DynamoDB, and foundational Lambda functions. | 4 Weeks | Core Backend operational, Auth/User Management completed. |
| Month 2: AI Integration | Integrate Claude Haiku 3 LLM on Amazon Bedrock (Stability AI), Replicate API, complete Image Processing functions. | 4 Weeks | Successful end-to-end AI image processing demo. |
| Month 3: Front-end & CI/CD | Develop UI/UX (Amplify/Next.js), finalize CI/CD pipelines, and configure Monitoring/Security (CloudWatch/WAF). | 4 Weeks | Full platform ready for user testing. |
| Month 4: Optimization & Go-Live | Perform performance testing (Stress Test), cost optimization, and Production deployment. | 4 Weeks | Go-Live (Official product launch). |
6. Cost Estimate (AWS Pricing Estimate)
Total Cost
- Monthly: $9.80
- Upfront: $0.00
- 12 Months: $117.60
Service Overview
| Service | Region | Monthly Cost | Upfront | 12-Month Cost | Notes |
|---|---|---|---|---|---|
| Amazon Route 53 | Asia Pacific (Singapore) | $0.50 | $0.00 | $6.00 | 1 Hosted Zone, 1 domain, 1 linked VPC |
| Amazon CloudFront | Asia Pacific (Singapore) | $0.00 | $0.00 | $0.00 | No specific configuration |
| AWS WAF | Asia Pacific (Singapore) | $6.00 | $0.00 | $72.00 | 1 Web ACL; 1 rule per ACL |
| AWS Amplify | Asia Pacific (Singapore) | $0.00 | $0.00 | $0.00 | Build instance: Standard (8GB/4vCPU); request duration 500ms |
| AWS CloudFormation | Asia Pacific (Singapore) | $0.00 | $0.00 | $0.00 | No extensions; no operations |
| Amazon API Gateway | Asia Pacific (Singapore) | $0.13 | $0.00 | $1.59 | 10k requests/month; WebSocket message 1KB; request size 30KB |
| AWS Lambda | Asia Pacific (Singapore) | $1.67 | $0.00 | $20.04 | 1 million invokes; x86; 512MB ephemeral storage |
| Amazon CloudWatch | Asia Pacific (Singapore) | $0.85 | $0.00 | $10.22 | 1 metric; 0.5GB logs in; 0.5GB logs to S3 |
| S3 Standard | Asia Pacific (Singapore) | $0.23 | $0.00 | $2.76 | 10GB storage; 20k PUT; 40k GET |
| DynamoDB On-Demand | Asia Pacific (Singapore) | $0.42 | $0.00 | $5.04 | 1GB storage; 1KB item; on-demand mode |
| Total (Estimate) | — | $9.80 | $0.00 | $117.60 | Based on AWS Pricing Calculator |
Metadata
- Currency: USD
- Locale: en_US
- Created On: 12/9/2025
- Share URL: AWS Calculator Link
- Legal Disclaimer: AWS Pricing Calculator provides estimates only; actual costs may vary based on usage.
AI Model Pricing
| Model | Resolution / Token Usage | Quality | Price per Request (USD) | Notes |
|---|---|---|---|---|
| Titan Image Generator v2 | < 512×512 | Standard | 0.008 | Fixed price per 1 image |
| Titan Image Generator v2 | < 512×512 | Premium | 0.01 | Fixed price per 1 image |
| Titan Image Generator v2 | > 1024×1024 | Standard | 0.01 | Fixed price per 1 image |
| Titan Image Generator v2 | > 1024×1024 | Premium | 0.012 | Fixed price per 1 image |
| Stable Diffusion 3.5 Large | Any | N/A | 0.08 | Fixed price per 1 image |
| Claude (text + image) | 40 input tokens + 1 image | N/A | 0.00195 | Price for 1 request including text and 1 image 1024×1024 |
Additional Options
| Mode | Augmentation | Price (USD) |
|---|---|---|
| text→img | no augment | 0.08 |
| text→img | with augment | 0.08195 |
| img→img | no augment | 0.012 |
| img→img | with augment | 0.094 |
7. Risk Assessment
| Risk | Impact | Probability | Mitigation |
|---|---|---|---|
| AI model inference latency | Medium | High | Use ElastiCache + SQS/SNS for async handling |
| Cost increase from model calls | High | Medium | Bedrock usage control, SageMaker autoscaling |
| CI/CD misconfigurations | Medium | Low | CloudFormation rollback policies |
| Security vulnerabilities | High | Medium | WAF, GuardDuty, PrivateLink, IAM least privilege |
| Third-party API dependency | Medium | Medium | Bedrock fallback to S3-stored inference results |
8. Expected Outcomes
Technical Outcomes:
- Complete serverless AI image generation workflow with secure CI/CD.
- Modular orchestration enabling rapid MLOps integration.
- Improved latency and reliability via caching and async workflows.
Long-Term Value:
- A foundation for AI as a Service (AIaaS) platform expansion.
- Ready-to-scale MLOps framework with automated retraining.
- Reusable cloud infrastructure for future AI products.