Building a scalable and compliant data pipeline with Fivetran

Because uncontrolled access patterns are difficult to scope, log, and evidence under audit.

HIPAA requires that access to PHI is purposeful, logged, and limited to the minimum necessary. Ad-hoc analytics queries against a production database rarely meet that bar; they're often broad, inconsistently logged, and run by users whose access was granted for operational rather than analytical purposes.

A dedicated ingestion pipeline through a controlled read replica creates a defined, auditable access path that maps cleanly to HIPAA's minimum necessary and audit control requirements.

Because structured incremental sync with clear state tracking reduces operational risk in regulated environments.

A custom pipeline can ingest data, but it requires ongoing maintenance, failure handling, and state management, all of which create operational overhead and failure modes that need to be monitored.

Fivetran provides incremental sync that resumes correctly after interruption, connector-level observability, and predictable ingestion behavior that can be scheduled and evidenced. For a team that needed analytics infrastructure without adding data engineering headcount, it was the right fit.

It creates a private network path between AWS services that never touches the public internet.

Without a VPC endpoint, traffic between your database and ingestion layer routes through the public internet by default, even if both are on AWS. For PHI data, that routing introduces exposure that HIPAA's transmission security requirements specifically address.

A VPC endpoint keeps the data path entirely within the AWS network, removing a category of transit risk and simplifying the network controls you need to document for compliance.

By serving analytics queries from a separate database instance that receives changes asynchronously from the primary.

A read replica maintains a copy of the production database that stays current through replication, but carries none of the write load and none of the application query traffic. Directing Fivetran at a read replica means ingestion jobs, which can be resource-intensive at a 12TB database scale run against an instance that exists specifically to absorb that load. Production query performance is unaffected regardless of ingestion schedule or data volume.

Source isolation, private transit, encrypted storage, controlled access, and an audit trail at every stage.

For this client, that meant: Fivetran pulling from a dedicated read replica rather than the production primary; private connectivity via AWS VPC Endpoint; TLS encryption in transit; AWS KMS encryption at rest in both Redshift and S3; IAM-scoped access to the destination data stores; and Fivetran's connector logs feeding into the platform's existing audit infrastructure.

Each of these controls addresses a specific HIPAA requirement and produces an artifact that the compliance team can reference. The architecture was designed so that every data movement decision could be explained and evidenced, not just implemented.