AI Engineering from Scratch: Six Images Lay Out the Scope
Rohit Ghumare posted a tweet containing only a title and six images, on the theme of AI Engineering from Scratch. There is no long-form explanation; the information is all in the images. It is useful for comparing against knowledge gaps, and also for non-technical readers to build a rough map.
Rohit Ghumare posted a tweet on X with just one line of body text: AI ENGINEERING FROM SCRATCH. It was followed by six images, with no long-form explanation attached. The link to the original tweet is at the end.
This kind of material is rare. Almost all the information is in the images, and the text only gives a theme. Let's be clear about the boundaries: I won't fill in a checklist on behalf of this tweet below, nor guess at content that isn't written in the images. The six images are placed here in their original order; the specific items are as shown in the images.
The term AI engineering is used very broadly now. Some people mean prompting, some mean retrieval-augmented generation, some mean model deployment, and others mean the whole pipeline of data, training, evaluation, and launch. The value of a from-scratch diagram is to let people see the scope first: don't expect to know how to do it all after looking at it. People with a development background can use it as a checklist to see whether what they lack is data pipelines, evaluation methods, inference optimization, or productization. Readers without a technical background can also use it to build a map and understand that this field is not just about being able to write prompts.
There is a common problem with looking at roadmaps: saving equals learning. You can save the images, but your hands need to move. Pick one node, find the smallest runnable project, and run it through once. That is more useful than putting the six images into your bookmarks.
There is no further textual explanation for the material. Rohit Ghumare's original tweet is here: link. The six images need to be enlarged one by one; don't just look at the thumbnails.
发布时间: 2026-11-18 02:27