Cloud Architecture & Engineering

By inventorying applications with their dependencies (databases, files, flows, licences), then choosing a treatment per application: rehost as is, replatform to a managed service, rebuild, or leave in place. The estimate includes double billing during the transition, licences and team time. We present it in waves, with the expected benefit of each, so the decision stays reversible.

Terraform or OpenTofu describe the infrastructure; you need to add reusable modules, code review, a plan run in CI before every apply and policies checked automatically (encryption, tagging, networking). What matters is not the tool but the rule: nothing is created in the console. We apply it to our own infrastructure.

In order of return: switch off what is not used (environments at night, orphaned resources), right-size on observed load, move tolerant workloads to spot instances or serverless, then commit on the stable baseline. And tie every euro to a team through tagging, otherwise the savings do not last three months.

A common foundation: object storage as the source of truth, an open table format, compute separated from storage, centralised identities and secrets, private networking between services. On that base, Snowflake, Databricks, BigQuery or native services plug in without lock-in. Data residency and disaster recovery constraints are set in the target design from the start.

By rehearsing it. We define recovery objectives (acceptable downtime and data loss) per service with you, verify that backups actually restore, then run a failure scenario on a representative environment. A recovery plan that has never been exercised is a hypothesis.