I do not write my articles with a keyboard. I speak into my phone and, within a few seconds, the text appears on my site. The words I am speaking now follow the same path: this conversation will become a blog post.
On my phone, a shortcut starts the recording. Once I finish speaking, the audio is sent to my server, where a two-stage process takes over. The first stage is a speech recognition model that converts my voice into text. The second stage is a language model that takes the raw Turkish transcript and produces professional English, including the title, description, tags, category, and date. I have given that model a detailed set of instructions, so it understands the tone, the format, and what to avoid.
Once the post is generated, it is committed directly to my code repository. The site rebuilds automatically, and I do not have to touch anything. Access is protected by a secret key, so only I can trigger the pipeline.
The architecture matters more than the novelty. Transcription and writing are two different problems, and splitting them keeps each one honest. A speech model optimised for accuracy does not have to also reason about structure, and a language model asked only to rewrite clean text does not have to guess at half-heard words. The secret key turns an open endpoint into a private one without the weight of a full auth system, and committing straight to the repository means the blog has no separate database to keep in sync: the post is the source of truth, and the deploy is the publish step.
The result is that writing this post required no keyboard and no editing session. I spoke for a few minutes while walking, and the article took care of itself.