Migrating a live course platform off LearnDash without breaking it

The challenge

The platform ran on LearnDash, and the goal was to replace it with a custom CRM backend as the source of truth for courses, modules, quizzes, categories and tags. The hard constraint was that the existing user interface logic could not break while this happened. Course and quiz data had to arrive in the new backend in a shape the front end already understood.

What we built

• A CRM backend API that owns course, module, quiz, category and tag data
• A batch sync architecture: initialise a sync, run modules in batches, and poll progress separately
• Sync-state persistence in MongoDB so a long-running sync survives interruption
• Quiz title and quiz question ingestion from LearnDash into the new data model
• Course-level aggregations for reporting
• A separate internal verification interface for confirming and debugging sync results

How we built it

We deliberately kept three systems separate: the backend and sync engine, an internal verification interface, and the end-user application. Mixing them is what turns a migration into an outage.

Syncing ran in batches rather than one long request, with progress tracked independently so the process could be observed without being blocked on a single response.

Verification pages were built for courses, course detail, module detail and sync status, so every synced record could be checked against its LearnDash origin before cutover.

Challenges we solved

• Long-running syncs versus request timeouts. A request can time out while backend processing continues happily, which looks like failure and is not. Splitting execution from progress reporting made the real state visible.
• An infinite loop in batch progression. When the course index failed to advance, batching looped. Making progression state explicit and persisted removed the ambiguity.
• Rendering data the front end did not expect. Feeding a new backend into existing React components surfaced errors from objects being rendered directly. Matching the new payload shape to what the UI already consumed resolved it at the contract level rather than patching components.
• Scope discipline. Migration hooks and compatibility layers that had crept into the wrong system were removed, keeping each of the three codebases responsible for one job.

Technology used

More work