Start by pointing Fivetran at each source; it handles the connectors, schedules, and schema drift so you only worry about warehouse tables. Pair it with dbt for any transforms you actually need, letting you version SQL instead of random Python. Airbyte is a solid open-source fallback if you want to self-host and tweak connectors. I’ve also leaned on DreamFactory for spinning up quick REST endpoints when the business wants the same data fed to microservices without another script. Stick to one ingestion layer, document lineage, and the nightmare fades.