By warming up the Azure Container Apps backend on the demo page-load, I cut the first-click analysis latency by roughly 5–10 seconds, depending on typing time.
Before: the first analysis took around 30s because of cold start. Now: Cloudflare Worker checks KV store on page load: >5min idle? Invisible prefetch to Azure Container Apps. Result: user gets faster response.
This text-analysis / AI chat wrapper started as a simple example project to practice key AZ-400 topics: deployment pipelines, observability, and edge-backend integration. I originally ran it on AKS, but the cost-structure wasn't feasible for a live demo, so I migrated it to Azure Container Apps (max 2 replicas), reusing the same container images except for Ollama.
Tech Stack: Cloudflare Workers + KV + Azure Container Apps (max 2 replicas) + Cloudflare AI
Demo: text-analysis-demo.koorevaar.com · Repo