Forum: Azure VM to StaticWebApp


Context & Motivation

Since my last update, I’ve significantly advanced my toolchain. Full integration of OpenAI into my workflow has dramatically boosted my efficiency:

  • ChatGPT 4.1 for code, text, and logic design
  • DALL·E for image generation
  • Codex for end-to-end coding

The following outlines how I leveraged these tools—especially Codex—to re-architect my Discourse-based forum infrastructure.


Problem with the Discourse Setup

As described in this earlier post, I deployed a Discourse forum via a Bitnami image on Azure. While functionally solid, it introduced disproportionate cost for a lightweight use case:

  • VM idle cost: ~30 €/month, even with optimizations
  • PostgreSQL performance overhead = non-reducible baseline
  • Difficult to scale horizontally without deep-diving into AKS/CosmosDB
Old VM Solution vs. New Static WebApp Approach

The visual above illustrates the transition from a bloated resource group to a single Static Web App—eliminating all compute and networking overhead.


New Approach: Static WebApp + Hugo

Given my positive experience with Hugo (which also powers this blog), I transitioned the entire forum to a Hugo-based Static Web App on Azure’s Hobby Tier:

  • 0 €/month running cost
  • Disqus for commentary
  • Room for growth via Azure Functions, if needed later

This simplified architecture now fully replaces the prior VM stack.


Codex-Powered Development Workflow

All development was executed after-hours, leveraging OpenAI Codex for GitHub integration:

  1. Repo linked to Codex.
  2. Prompted Codex with functionality requirements.
  3. Codex auto-generated code + pull requests.
  4. I reviewed & merged ~40 PRs.
Workflow with OpenAI Codex incl. Pull Request

The process exceeded expectations—Codex even migrated a complex C# code-generation module to Python without intervention.


Early Result Preview

After a few dozen iterations and Codex contributions, the new forum frontend was live. The layout was minimal, but fully functional and lightweight:

List View of the new Site

Feature Breakdown

The final system integrates several modern frontend features, all implemented with Codex-driven PRs:

Finished Site after >40 Iterations / Codex PR's
  • Dark mode toggle
  • Client-side search via lunr.js
  • Web Share API
  • Fully responsive layout
  • Collapsible sidebar navigation
  • GitHub Actions pipeline
  • Kit.com newsletter signup (also now live on this blog)
  • Disqus as a lightweight pseudo-forum solution
  • Two live entrypoints:
    • reference.wtf-model.com
    • forum.wtf-model.com

Conclusion

This migration not only eliminated a recurring infrastructure cost but also redefined my delivery workflow through deep AI integration.

OpenAI Codex demonstrated high stability even under complex build cycles. With this experience, I’m confident in treating it as a dependable contributor to future mid-sized projects.

comments powered by Disqus