Self-Hosting Your Own AI Model on Azure
Run a real language model behind your app for very little a month, using a container that scales to zero, and understand what end to end encrypted really means once an AI is in the loop.
skyel
skyel publishes technical writing, product previews, and professional portfolio pages. The public site is designed so people and crawlers can understand the work before the Flutter experience loads.
The public routes on skyel.net are intentionally narrow. They introduce the people behind the work, publish technical essays, explain the current product direction, and expose enough real copy for search engines to understand each destination without depending on client-side rendering.
That public layer matters because the site is also a Flutter web application. Without route-level prerendered content, crawlers only see a JavaScript shell and thin metadata. The prerendered pages fix that by shipping meaningful HTML for each public route before the application takes over.
skyel vault is the encrypted password manager on the site. closharing is the private media and communication workspace built around sources that people already trust. Both public routes explain the product model, the current development state, and the boundaries of the workflow before any authentication step.
The blog complements those products with engineering essays and technical writeups that explain how the systems are built, what tradeoffs shaped them, and where the work is heading next.
Run a real language model behind your app for very little a month, using a container that scales to zero, and understand what end to end encrypted really means once an AI is in the loop.
Bayesian inference begins with a prior. The quality of the posterior is bounded by the quality of the prior. So is the quality of every answer you accept from a language model.
No laptop, no Android Studio, no cloud VM. Turn your phone into a full Android build environment using Termux, proot, and QEMU.