AIOS Platform

Overview

The production platform I run day to day: a multi-client AI system on Azure Kubernetes Service, with end-to-end observability, alerting, and incident response owned by me.

The problem

A multi-client AI platform running on 4 AKS clusters needed a single owner for reliability: someone watching the dashboards, tuning the alerts, and keeping the infrastructure bill under control as usage grew.

Key details

• 4 production AKS clusters plus PostgreSQL, Functions, Key Vault, and Storage on Azure.

• 40+ Grafana Cloud dashboards on Prometheus, Loki, Tempo, and Application Insights covering logs, calls, LLM usage, and billing.

• 30+ Grafana alert rules routed to Slack and email.

• Self-hosted Trieve RAG stack on AKS: Qdrant, embedding and reranker models, Keycloak SSO, ClickHouse.

• n8n workflow automation running on a dedicated VM for platform integrations and pipelines.

Tech used

Outcomes

• Reduced monitoring spend roughly 71%, from about $1,090/mo to about $316/mo.

• 40+ dashboards and 30+ alert rules give full visibility into logs, calls, and LLM usage.