GitIngest (https://gitingest.com/) is an awesome tool that allows you to batch together files from a GitHub repo in order to use them with AI tools. This lesson walks through use Git Ingest against a repository that's too large for an AI, so it configures options to narrow down to only the files that you need for your task. Then we'll use Gemini's latest Flash 2.0 model to generate a new project based on the examples we grabbed from the GitHub repo.
Transcript
[00:00] So git ingest is going to allow you to take a repository, let's try angular's, and then grab all the files that you need to drop into an AI. So while we can generate a file from the entire angular repository, this would end up in the millions and millions of tokens range. So we're going to only include things from the packages, examples, and only all of the TypeScript files. This way we'll only get the files under packages, examples, and we'll have all the examples from here. So now that we've limited the number of files, we'll hit ingest.
[00:37] And you'll see the directory structure, but we're going to download and then go over to AI Studio, and from our downloads we'll grab the file we downloaded, just drop it in here. You'll see that came out to about 60, 000 tokens, well within the limit. And we can prompt for, based on the included examples, please generate a minimalistic Twitter clone and paste that in. And then once we run this, we'll know that the code that it generates is going to be finely tuned on the most recent code from Angular, which can be extremely helpful because models are often out of date from what the most recent code is. And after 26 seconds you can see all of the code it generated and all of this should be based much more closely on what would be considered best practices based on the Angular examples rather than what the model is trained on.