Safwene Ben Aich

For over 15 years, Believe has been implementing digital distribution and marketing solutions for independent labels and artists in 50 countries worldwide with a turnover of €988.8 million in 2024.

I joined the SRE team (10 engineers including two Heads of SRE) with the primary mission of accelerating the migration of the platform (on-premise & AWS) to GCP, providing support, build, and task force.

I performed the following tasks:

- Study and deployment of Spot VMs instances in non-prod environments -> €700 in monthly savings (~ -30% of off-prod GKE cost).

- Implementation of Datadog Dashboards for Spot nodes (preemption alerts, availability monitoring).

- Study and migration plan for the managed Apache Airflow solution on AWS (MWAA) to GCP (Cloud Composer 3).

- Identification of migration challenges and critical dependencies (Databricks, S3, Lambda, AWS Secret Manager, etc.).

- Development of a comprehensive action plan (services, resource sizing, IAM/RBAC access, network, storage, IaC, CI/CD).

- Development of dedicated Terraform modules with contributions to the official terraform-google-modules/composer module.

- Implementation of CI/CD pipelines with Gitlab-ci.

Here are the main results achieved:

- 30% reduction in GKE costs for non-production environments.

- Successful migration of the royalty distribution infrastructure (Apache Airflow) from AWS to GCP.

Technical context:

- AWS

- GCP

- Vault

- Terraform

- gitlab-ci

- Kubernetes

- ArgoCD

- Terragrunt

- Airflow

- Datadog

- Git

- SRE

- incident.io

Doctolib is the European leader in digital health, with over 300,000 healthcare professionals and 70 million monthly users. Founded in 2013, this scale-up has raised over €500 million to support its expansion.

I joined the Data Engineering team, composed of 3 data engineers, an engineering manager, and a product manager, to lead the implementation of a distributed architecture for an event capture API on AWS.

I performed the following tasks:

- Study of the existing AWS architecture, identifying potential areas for improvement and urgent issues to resolve (e.g., SPOFs & bottlenecks).

Proposed action plan:

- Development of a technical architecture document as a source of truth for the existing setup.

- Re-architecture of the networking part (VPC, VPC Endpoints, subnets, Security groups, load balancing, routing, etc.).

- Splitting the ECS service into multiple services to resolve SPOF and timeout issues.

- Migration of the compute part from AWS ECS to AWS EKS.

- Development of Terraform modules for the various services that will run on ECS and then EKS.

- Development and implementation of the Kubernetes workload on EKS (deployment, service, configMaps, CRD resources, etc.).

- Implementation of the CI/CD part with Github Actions (Terraform, Kubernetes with Helm).

Here are the main results achieved:

- Resolution of the API timeout/SPOF issue, no more data loss.

- Reduction in infrastructure cost due to proper resource sizing ≃ -20%.

- Simplification of deployment and CI/CD flow following migration to Kubernetes.

- Facilitated migration of the application to GCP due to fewer dependencies on AWS solutions.

Technical context:

- AWS ECS

- AWS ALB

- AWS Lambda

- AWS Kinesis

- AWS ECS

- AWS S3

- Kubernetes

- Terraform

- Redshift

- Airflow

- Python

- JIRA

- Scrum

- CI/CD

- github-actions

- Helm

- Vault

- TFE

- DevOps

- Git

- Datadog

- ELK

- Review of the N8N solution architecture

- Study of AWS -> GCP cloud migration

- Incident support