Now that I have put together my little RAG project (little but functional, more than a mere toy demo) it led to another idea. The abstract vector database (embedding) that represents my answers can be visualized, well, sort of, in a two-dimensional representation, and I built just that: an interactive visualization of all my Quora answers.
It is very educational to explore, how the embedding model managed to cluster answers by semantics. As a kind of a trivial example, there is a little “cat archipelago” in the upper right quadrant: several of my non-physics answers related to cats can be found in this corner. Elsewhere there is, for instance, a cluster of some of my French-language answers.
Anyhow, feel free to take a look. It’s fun. Unlike the RAG engine itself, exploring this map does not even consume any significant computing (GPU) resources on my server.