After a few years of keeping KD7 healthy on an end-of-life stack, I wrote up a plan to modernize it from the ground up. It didn’t get the go-ahead at Apti, but I couldn’t stop thinking about it. So I kept building it on my own time.
KD8 is a full rebuild of a multi-tenant learning platform: Laravel 11 on PHP 8.3 instead of PHP 5.6 and Phalcon, an Angular 19+ Nx monorepo instead of AngularJS 1.7, and an isolated server for every client instead of shared ones.
Originally proposed as a platform modernization initiative at Apti; continued independently as a personal project after it did not receive organizational buy-in.
Every layer of the stack gets a fresh start:
The biggest architectural change is isolation. In KD7, every client shares the same servers and the separation happens in code. In KD8, Terraform provisions a separate EC2 instance for each client, and cloud-init bootstraps it from scratch with nobody logging in to set it up. Data isolation rolls out in phases, so the move can happen gradually instead of in one terrifying weekend.
Getting clients there is its own project, so I built custom tooling that pulls legacy KD7 data, transforms it for the new schema and loads it into KD8.
Accounts move through clear states (pending, active, deactivated), gated by email activation. That’s a direct lesson from KD7, where I once found a deactivated user could reactivate themselves through a password reset. In KD8, the account lifecycle is designed in from the start instead of patched on later.
Each user belongs to exactly one category, and can join any number of groups for course assignment.
GitLab CI tests, builds and deploys every change, so the new platform starts life with the delivery habits the old one had to have bolted on later.
I use Claude to dig through how the legacy platform actually behaves and to scaffold new work, which keeps a one-person rebuild moving at a steady clip. The architecture and the plan are mine; the AI helps me get there faster.
Maintaining KD7 taught me where it hurts. KD8 is my chance to design those pain points out instead of working around them.