Build Log
Notes on building a Kubernetes-native AI agent platform on Google Cloud.
Why I Forked Sympozium for 100% Google Cloud
The original platform supports OpenAI, Anthropic, Azure, and Ollama. I stripped it all out and rebuilt for GKE + Gemini. Here's why a single-cloud bet makes sense for Kubernetes-native AI.
Read more →Hybrid Inference on GKE: Vertex AI + Gemma on TPU Nodes
Running Gemma 3 (27B) on TPU v5e nodes alongside Vertex AI Gemini creates a cost/capability spectrum. The node-probe DaemonSet discovers local engines automatically.
Read more →Two Observability Paths: Cloud Monitoring vs Prometheus/Grafana
Not every team wants Cloud Monitoring. We added Google Managed Prometheus support with PodMonitoring CRDs, alerting rules, and a 14-panel Grafana dashboard.
Read more →Keyless Auth with Workload Identity Federation
No service account JSON keys anywhere in this platform. Every pod authenticates to GCP via Workload Identity — Vertex AI, Pub/Sub, Secret Manager, Firestore, all of it.
Read more →Building a Google Chat Bot with Cards v2 and Slash Commands
The Google Chat channel implementation: Cards v2 responses, /ask, /status, /run, /help slash commands, threading, typing indicators, and interactive retry buttons.
Read more →