Webinar Migration Pipeline

Context

Every webinar/masterclass run generates up to seven distinct follow-up email variants — no-shows, all-registrants, attendees who dropped off early, attendees who saw the offer, timed reminders before the event, timed sends from sign-up, and existing customers. When the platform's email system changed, all of this legacy content needed to move to the new system — one email at a time, by hand.

The problem

Rebuilding these emails manually meant three repetitive, error-prone steps for every single template: rewriting the HTML to match production standards, manually converting every legacy merge variable ({{ webinar_time }}, {{ webinar_replay_link }}, and a dozen others) into the new platform's templating syntax, and re-inserting the correct header, footer, and view-in-browser blocks by hand. Multiply that across every webinar and every email type, and it becomes a standing bottleneck — slow, inconsistent, and easy to get subtly wrong (a single missed variable breaks personalization for an entire send).

What I built

A Python migration pipeline that takes a raw data export and turns it into fully production-ready, live templates with a single command:

1. Each email's raw HTML is passed to an LLM agent that handles content cleanup and rewriting, following a defined quality standard
2. A mapping layer automatically converts every legacy variable to the new platform's correct syntax — no manual find-and-replace
3. The processed content is wrapped in the platform's production scaffold: correct table structure, mobile-safe styling, and the required header/footer/view-in-browser blocks
4. The finished template is created or updated directly via API

The pipeline is idempotent by design — re-running it never creates duplicates, it just safely updates what's already there. It also supports dry-run and single-record test modes, so changes can be verified before anything goes live, and every run produces a full audit trail (per-row status, flagged issues, template IDs) rather than relying on someone manually checking each email.

Stack