we need full text indexes first!

algorithms kinda depend on that

it's also a good use case for LLMs to take those fulltext searches and use their semantic juju to filter and sort results by relevance, there are ways to do this with simpler stuff but feeding an LLM a set of results to rank them by relevance is basically what they were born for. i've been using LLMs more in my programming work and what they are really good at is sifting through a lot of data and picking out relevant stuff. they are not so smart at writing code because of the causality relations between processes and ontology of the data. this is why there is these "agent" things, which basically use similar principles as language compiler state machines to define a procedure, and the LLM creates plans, and then evolves them as it acquires more input from the code.

anyway, yeah. we need full text indexes first. DVMs should not exist - the relay should have this facility built into it, and then a worker that takes results and sends them to an LLM to filter and sort them.