“I’ve recently started making RAG apps using ollama and python, while they work I wouldn’t call them perfect. So I’d love to see what a good RAG application has under the hood. Let me know how you designed your pipeline (which splitting, which vector db, which embedding model, prompting techniques, other optimization techniques etc) and if you’re okay with sharing your git repo that’s an added bonus”
With enough people in the comments, one of the commenters who wrote details about their approach but without the source code ended up open-sourcing their repository. In these types of posts, it isn’t the OP who takes the lead in explaining, or providing resources, it is the commenters.
One other member shares that they use RAG to maintain track of novel world building, such that if a character learns something or changes information then the RAG is used to avoid breaking common sense of the character.
Many share their repos and their approach (1, 2, 3), but some share papers that they published on the topic of RAG.
One shared a YouTube video of their work which tries to use RAG for hiring. Another created a list of good RAG projects.