ALTO device platform

01The problem

Every IPTV device on the ALTO platform keeps a live connection for commands, status and notifications. At 100,000+ devices, the backend has to hold every socket open, route messages to the right device and survive spikes without dropping events.

On top of that, the original backend was a monolith: one slow path could drag the whole platform's latency up.

02Approach

I designed the real-time device management system on NestJS with Socket.io for the device connections and MongoDB for device state.

Notifications run through a WebSocket and queue-based event pipeline that delivers millions of events a day with at-least-once semantics, so a slow consumer delays a message instead of losing it.

We split the monolith into NestJS services on Docker and Kubernetes. Isolating failure domains cut p95 latency by about 40%.

03Architecture

Devices connect to NestJS Socket.io gateways; events flow through queues to NestJS services backed by MongoDB.

A schema-first GraphQL API is the single layer for web, mobile and partner integrations.

Everything ships through Jenkins CI/CD to Kubernetes on AWS EKS, with monitoring and incident response owned by the team I led (7+ engineers).

04Results

100,000+ concurrent device connections, millions of events a day, and p95 latency down ~40% after the services split. The same device telemetry later fed the ALTO AI agent and MCP server.