I'm working on a project that involves numerical, time-series data on the order of 10s of GB. I was wondering if vector embeddings would be the way to go if I wanted to accurately retrieve and perform analysis on this data via LLM. Most of the resources I've run into revolve around embedding text, images, and other forms of media, so I'm not sure if the temporal relationships between hundreds/thousands of time-series can be represented effectively the same way.

Has anyone here had experience with doing this? Or any libraries or projects you could point me towards? I've found a few projects like Time-LLM that attempt to forecast time-series, but not query or analyze them.

Any guidance would be very appreciated, thanks!