The challenge
The company runs a SaaS platform that helps manufacturers get their connected products certified against cybersecurity standards. Its customers are security-conscious by definition, so the platform's own infrastructure had to hold up to the same scrutiny: data in the EU, encryption everywhere, audit trails and a clear answer to "what happens if something fails?".
The application had grown up on Heroku. That was fine for the first version, but it no longer fitted: more services, an AI-assisted assessment stack with vector search, several customer-specific environments, and growing questions from customers about where data lives and how it is protected. Releases were manual and slow, and there was no single place to see whether the platform was healthy.
What we did
- A new home on AWS in Paris. An EKS cluster in eu-west-3 spread across three availability zones, with managed PostgreSQL (with pgvector for the AI features) and Redis in Multi-AZ mode. Everything is defined in Terraform and Helm, so environments can be rebuilt from code.
- The move off Heroku. Every service was containerised and migrated to the cluster, one environment at a time, so customers kept working throughout.
- Delivery that runs itself. GitHub Actions builds each service, pushes the image to ECR and deploys it with Helm. Development and staging update automatically in about ten minutes after a merge; production deploys stay a deliberate, manual step.
- Security built into the platform. Customer-managed KMS keys for encryption at rest, TLS 1.2+ with HSTS at the edge, workloads in private subnets, per-service AWS permissions through IAM roles for service accounts instead of shared keys, and a written key-rotation policy.
- Monitoring the team can act on. Prometheus, Grafana, Loki and Alertmanager, with alert rules grouped by severity, external uptime and TLS-certificate-expiry probes, and Sentry for application errors.
- Backups that cover every environment. Automated daily database dumps and configuration backups to S3 across all environments, on top of daily managed snapshots with point-in-time recovery.
- GDPR paperwork that matches reality. Records of processing, a sub-processor list and retention and vulnerability-disclosure policies, written from how the system actually works.
Results
- All production and customer environments run on EKS in an EU region, defined in code.
- Changes reach staging in about ten minutes after merge, instead of waiting for a manual release.
- Dozens of alert rules and external probes mean problems, including expiring certificates, show up before customers notice.
- Daily, automated backups across twelve environments, plus point-in-time recovery for the databases.
- Right-sizing the non-production node group cut that part of the compute bill by a third.
- Answers to customer security questionnaires now point to real controls and documents.
Stack
AWS EKS (Paris, 3 AZs) · Terraform · Helm · GitHub Actions · Amazon ECR · RDS PostgreSQL + pgvector · ElastiCache (Multi-AZ) · KMS · Prometheus · Grafana · Loki · Alertmanager · Sentry
In their words
LL Cloud transformed our deployment pipeline and reduced our release cycle from weeks to hours. Their deep understanding of security requirements in our industry made them the ideal partner for our infrastructure.
— CTO, cybersecurity certification platform