The author also mentions that the basic steps provided are usually sufficient, but there are parameters in these steps that can be adjusted. These include how you chunk (chunk size and rules), how you embed (model and size), and how you query (metrics used). The author emphasizes that the second step is crucial for the quality of results as it determines what can be used in the other steps, particularly the embedding comparison metric that defines relevance.
Key takeaways:
- A mature search product likely improves upon a single embedding method by using a combination of indexes, queries, and rankings.
- These systems often attach more data and linkage to an item than just the content's embedding, and perform multiple queries simultaneously.
- Parameters in the basic steps of chunking, embedding, and querying can be adjusted for improved results.
- The choice of embedding model and size is crucial as it determines what is available for use in other steps, especially the embedding comparison metric that defines relevance.