Backends
that
stay up

Subscription billing for SaaS

Problem

Invoicing was kept in spreadsheets. As soon as plans, proration and refunds appeared, every month ended with manual reconciliation and mistakes that reached customers.

My role

Sole backend developer: domain model, REST API, payment-provider webhooks, containerisation and deployment.

Hard part

Payment webhooks arrive out of order and more than once — a naive handler charges twice. I built an idempotent event log with a deduplication key and a retry queue on RabbitMQ, so a repeated callback is recorded but never billed again.

• PHP 8.4
• Symfony
• PostgreSQL
• Docker
• RabbitMQ

CRM for small teams

Problem

A small sales team tracked deals in chat and notes. Nothing was searchable, nobody could tell which deals had gone quiet, and handover between people lost context.

My role

Full stack: Laravel API and admin, Vue 3 interface, role model, Docker Compose setup for local work and production.

Hard part

The deals list grew to thousands of rows with related contacts, tasks and stages — classic N+1, the page took seconds. Eager loading plus a materialised summary table brought it under 100 ms and kept it there as the data grew.

• PHP 8.3
• Laravel
• Vue 3
• MySQL
• Docker

Notification bus over RabbitMQ

Problem

Several services each sent their own Telegram messages. Rate limits were hit, failed sends disappeared silently, and every new service duplicated the same code.

My role

Designed and built the service: message contract, queue topology, templates, retries and monitoring.

Hard part

Telegram limits the send rate and returns errors that must be treated differently — some are worth retrying, some are not. I added exponential backoff with a dead-letter queue, so a broken message stops eating the queue instead of blocking everything behind it.

• PHP 8.3
• Symfony
• RabbitMQ
• Redis
• Docker