TaskStation
A multi-service platform in Bangladesh. We built an AWS-native DevOps setup for it.
The facts
The result
20 ms at p95Purchase latency, 95th percentile, even at peak load.
The challenge
- Deployments were manual, so new features were slow to reach users.
- Releases behaved differently between QA and production.
- No clear branching strategy for parallel features, bug fixes and releases.
- Services (Integration, Platform, Auth, Report) ran across three availability zones and had to be deployed in step.
- Little automated testing or rollback, so human error was a constant risk.
- Weak visibility across distributed services, so issues were found late.
- Sensitive configuration and access needed tighter control.
What we did
- Set up a GitHub branching model: feature, development and release branches.
- Built CodeBuild pipelines that build and test on every commit.
- Created separate QA and production environments.
- Provisioned VPCs, subnets, ALBs, ECS clusters and RDS with CloudFormation.
- Automated delivery to every environment with CodePipeline and CodeDeploy.
- Ran unit and integration tests in the pipeline.
- Added CloudWatch for metrics, logs and alerts.
- Moved secrets to Parameter Store and enforced least-privilege IAM.
How it was built
- Source and build
- GitHub
- AWS CodeBuild
- Deliver
- AWS CodePipeline
- AWS CodeDeploy
- Amazon ECR
- Run
- Amazon ECS across three availability zones
- Application Load Balancer
- Services: Auth, Platform, Integration, Report, Admin
- Data and messaging
- Amazon RDS for PostgreSQL
- Amazon S3
- Amazon SQS
- Infrastructure as code
- AWS CloudFormation
- Amazon CloudFront
- Amazon Route 53
- See
- Amazon CloudWatch logs, metrics and alerts
- Protect
- IAM least privilege
- AWS Systems Manager Parameter Store
- Encryption policies
- 90-day secrets rotation
What came of it
- Purchase latencyp95 under 20 ms at peak load
- Database uptime99.97% to 99.99%, month over month. A measured result, not a service commitment.
- Data lossNone recorded. Point-in-time recovery tests passed.
- CostAbout 60–65% lower infrastructure cost
- Security100% encryption compliance. No major or high-severity audit findings.
- Change failures3–5% of deployments
- Recovery drillsRun quarterly
- Customer satisfaction4.8 out of 5
More in Cloud Build & Launch
- eMarketing 23, for banksBanking, bulk email, BangladeshBuild, on AWS1 million emails in 3 hours (opens in a new tab)
- European biopharmaceutical companyBiopharmaceuticals, Several European countriesBuild, on AWSMore registrations and attendance (opens in a new tab)
- European tour bus platformBus travel marketplace, EuropeBuild, on AWSFaster content delivery (opens in a new tab)