ConnectCRM started from a concrete observation: Nigerian sales and marketing teams were either using spreadsheets or paying for expensive Western CRM platforms that had no knowledge of Paystack, Termii, or SMSLive247. I was brought in to architect and build the entire backend from scratch — a multi-tenant SaaS CRM that could be white-labeled for any organization while keeping each tenant's data completely isolated.

The backbone of the system is a 24-module Domain-Driven NestJS application backed by PostgreSQL, Redis, and Meilisearch. I organized every feature into its own domain module — each with a controller, service, entity, and DTOs — so the codebase structure mirrors the business domain. Multi-tenancy is enforced at every layer: the JWT payload carries an organization_id, a middleware re-injects it into the request context on every call, and every single TypeORM query is hard-filtered to that org.

One of the most satisfying design decisions was building a unified polymorphic system for notes, tasks, files, activity logs, and tags. Instead of creating separate tables for 'lead notes' and 'deal notes' and 'contact notes', I used a single table per concern with an entity_type + entity_id discriminator pair. Every current and future entity type gets all five capabilities automatically, and adding a new entity to the CRM requires zero schema migrations.

The campaign engine was the most complex system in the project. It's a 4-step state machine supporting both email and SMS channels. I built a PersonalizationService that replaces Handlebars-style tokens per recipient, and a LinkProcessingService that rewrites every URL in the content to a short-code redirect. All sends are async through Bull queues backed by Redis, so launching a campaign to thousands of contacts never blocks the API.