I've been working recently with PrivateGPT and have build content scrapers to pull articles of reference to load. Currently my project wipes PrivateGPT each day to load and summarize the prior day's set of articles. This seems like the best way to get a true representation of the content from the prior day.
Ideally though I'd really like to be able to prompt Chat about what was written on a certain day or period of time (e.g., what was reported last week about XYZ?). I'm having a hard time though conceptually visualizing how this is even possible. I'll start by saying that the first line of every article I prepend something to the effect of "On August 6th 2023, John Smith from THIS SOURCE reported from THIS PLACE that...". Looking at the source chunks that come back it doesn't initially seem that the content being retrieved from the underlying vector database is able to reference a date time (though it of course references the source chunk of text that drove its answer). You could extend this to perhaps even adding a geospatial component (e.g., please summarize what was reported last week for THIS PLACE).
Any explanation or ideally reference to online sources to dive deeper into how vector databases (i.e., Chroma) as a tool set coupled with LangChain drive LLMs would be appreciated. Apologies in advance if my question wasn't clear or is uninformed as I'm just starting out and at end of day really want to understand what I'm building.
=-Dave