Success Story
Clostech migrates its AI product for e-commerce from Azure to AWS in six weeks
At a glance
Clostech is an AI startup for e-commerce: its app lets people try on clothes virtually with its own AI model. It ran on Azure with high costs and billing issues that caused temporary account suspensions. Craftech migrated the product to AWS in six weeks, ahead of key commercial events. Today the AI server runs with no risk of billing-related interruptions, on a scalable foundation with automated pipelines.
6 weeks
to complete the migration from Azure to AWS
Clostech is an AI startup for e-commerce. Its app lets people try on clothes virtually by uploading a photo and overlaying garments with its own AI model, retrained from open-source models. Its goal is to reduce returns and increase conversion for online retailers.
Solution implemented
- Amazon EC2 with GPUs for the AI workloads
- Dockerfiles for the backend and AI, deployed in containers managed by Amazon ECS
- Amazon RDS for PostgreSQL for the relational databases
- Amazon S3 and Amazon CloudFront to store and deliver images with low latency
- Amazon Route 53 and an Application Load Balancer for DNS and high availability
- Secure, centralized access with VPN and AWS IAM Identity Center
- CI/CD with Bitbucket Pipelines to ECS and CloudFront, AWS Secrets Manager for credentials and a VPC with private subnets
Business Challenge
Clostech ran on Azure with several limitations. Its compute requirements were intensive: it needed high-capacity GPU instances to process images in real time. Costs were high, because the 24-core, 98 GB RAM virtual machines were inefficient. Dependencies were spread across the frontend and backend on Azure, PostgreSQL databases and storage in Blob Storage. On top of that, billing issues caused temporary account suspensions. The migration had to be completed in a month and a half, without affecting product availability during critical sales dates.
Solution
Craftech moved the AI workloads to Amazon EC2 with GPUs and created Dockerfiles for the backend and the AI, deployed in containers managed by Amazon ECS. The relational databases moved to Amazon RDS for PostgreSQL, and images are stored and delivered with low latency through Amazon S3 and Amazon CloudFront. Amazon Route 53 and an Application Load Balancer handle DNS and high availability. Access was centralized and secured with VPN and AWS IAM Identity Center, and CI/CD was built with Bitbucket Pipelines to ECS and CloudFront, with AWS Secrets Manager for credentials and a VPC with private subnets.
Outcomes
- Migration completed in six weeks, ahead of key commercial events
- More agile and scalable infrastructure, with auto scaling for demand peaks
- Cost optimization with AWS-native services and storage on S3
- AI server availability with no risk of billing-related interruptions
- A solid foundation to evolve the product, with automated pipelines and modern security practices
Next steps
- Expose the AI API through a load balancer
- Migrate critical workloads from Amazon ECS to Amazon EKS
Stack & technologies
