Architected the platform that ran a Belgian real-estate business for five years.

The problem

The founder had a working but fragile MVP: a Belgian real-estate lead-gen platform that ingested leads from portals and resold them to brokers. It worked enough to have customers. It also broke on a weekly cadence, ran on a single web server with no queue infrastructure, billed manually, and stored its source of truth across a database and a Google Sheet maintained by a part-time virtual assistant.

He didn't need a junior dev to ship features. He needed someone who could take operational responsibility for the platform end-to-end while he ran the business.

The approach

A retainer-shaped engagement from the start. No fixed scope, no milestone document. The deal was simple: I owned the platform technically; he owned the business and the customer relationships. We talked async, met monthly, made decisions in writing.

The first six weeks were stabilisation - moving the platform onto proper infrastructure, introducing queues and Horizon, getting the database schema honest, eliminating the most embarrassing security holes.

After that the work became continuous: features as the business needed them, infrastructure work as scale demanded, refactor as the codebase revealed where it hurt.

The system built

Over five years the platform grew into a full operational stack for the business:

• Lead ingest pipelines for six-plus portal sources, each with its own format quirks, change-without-notice habits, and authentication regime. Idempotent receivers, error queues, alerts when a feed fell silent.
• Lead scoring and routing. Rules-based logic encoded in plain PHP - geography, broker capacity, language, price band, time of day. Eventually augmented with targeted LLM calls for first-pass qualification of free-text lead notes.
• CRM integrations with the platforms the brokers actually used, plus an internal CRM for the brokers who didn't have one.
• Custom billing engine. Usage-based, with overages, with reconciliation, with audit-grade exports. Replaced a spreadsheet and an inbox.
• Lifecycle email automation - onboarding, broker performance reviews, prospect nurture, dunning.
• Internal operator dashboards for the founder, the support team, and the brokers themselves. Different views, different permissions, same source of truth.
• Monitoring, queue health, alerting that woke me up exactly when it should have and never otherwise.

The codebase stayed Laravel throughout. The orchestration stayed visible - no no-code spaghetti graph, no SaaS lock-in, no platform we couldn't migrate off if we needed to.

The outcome

For five years the platform processed thousands of leads per month, ran the company's daily operations, and stayed online while the founder ran the business side. Three full-time employees and a network of brokers depended on it working. It worked.

The engagement wound down in early 2026 when the founder pivoted budget from operational infrastructure into go-to-market. That's a healthy reason for a retainer to end - the technical work had reached a place where less of it was the right answer. The platform is still running. I'm still on residual maintenance work as it transitions.

It is the proof point for every page on this site: this is what a five-year embedded operator engagement actually looks like.