What Should I Know Before Starting a RAG Project?

It sounds like you're getting a bit confused because RAG involves three separate concepts that are often stacked together. Start with embeddings by generating them for different sentences and manually calculating cosine similarity—this will help clarify what they are. Then, move to vector search by exploring how to implement similarity searches. Finally, piece everything together to understand RAG—it’s just about injecting results from your previous findings into a prompt. I even built a simple RAG implementation you can check out on npm or GitHub to see the theory in action!

To really grasp the fundamentals, consider starting with a college-level machine learning course. However, just taking those courses may not address everything you need, especially regarding vector databases, which are quite new and not always covered in textbooks. Don't shy away from experimenting with building projects even if you're not an expert yet—many people do just that! I recommend combining practical experience with some shorter online courses to get relevant theories. A good starting point could be online courses that focus on the essentials without overwhelming you. And if you want this as a career, then definitely take those math-heavy college classes!

The best way to learn is definitely through practice! If you're serious, take the time to read papers on platforms like arXiv, use notebooklm to better grasp concepts, and try out small projects with different RAG methods. I spent too much time searching for the best learning path instead of just doing and learning through projects. And don’t forget, IBM has some great resources related to LLMs that could be really helpful!