Senior Platform Engineer
Ingka Group (IKEA)Utsikt
Part of the team that built and maintained the platform and infrastructure for the AI/ML part of Utsikt, where a genetic algorithm was trained on large volumes of sales data spanning many countries and markets to optimize pricing. My focus was making the model trainable, repeatable and observable in production, running containerized Linux workloads on Cloud Run, distributing the load so training could scale across markets, reading training data from BigQuery and persisting every training result in Cloud SQL. The platform was defined as infrastructure as code in Terraform and rolled out GitOps style through CI/CD pipelines in GitHub Actions, with the repository as the source of truth, so the whole training environment could be recreated from git. I maintained and developed the platform in production, instrumenting the training runs with Cloud Monitoring, Cloud Logging and Cloud Trace, handling troubleshooting and incident response when jobs failed or degraded, and tuning how workloads were sized and scheduled to keep cloud cost proportional to the value of each run. Security was part of the platform rather than an afterthought, with secrets in Secret Manager, least privilege service accounts and policy as code. Quality was guarded by automated testing in CI, with snapshot tests to catch any unforeseen changes in training data and results, Pytest for unit tests and Playwright for end-to-end tests. I worked closely with the data science team to take models from experiment to production.
- MLOps
- Distributed training
- Platform engineering
- Python
- Infrastructure as Code (Terraform)
- GitOps
- CI/CD (GitHub Actions)
- Docker
- Linux
- Cloud Run
- Cloud SQL
- BigQuery
- Bash
- GCP
- Observability (Cloud Monitoring
- Logging
- Trace)
- Incident response
- Secrets management
- Least privilege IAM
- Policy as code
- Cost optimization
- Automated testing
- Snapshot testing
- Pytest
- Playwright
- GitHub Copilot
