Epic: llmoso supports single-stage Information Retrieval

[ryscheng]

What is it?

Basic information retrieval is just asking questions that we think can be solved with a single query to OSO.

In-scope examples:

• Which projects have a deployer address associated with it? (select from artifacts_by_project, join with projects_v1)
• Tell me which projects in Season 7 have the highest star count? (select from projects_by_collection, join with collections_v1, join with timeseries_metrics_by_project, order by)

Out-of-scope example:

• What percentage of Season 7 projects have a deployer address associated with it?

I'm trying to distinguish between a single-stage translation and multi-stage reasoning, where in the out-of-scope example, the agent needs to know to make multiple queries, how they relate to each other, and how to combine the results (divide). Since we don't have support for CTEs in the semantic layer, we treat this as multi-stage

[linear[bot]]

OSO-764 Epic: llmoso supports single-stage Information Retrieval

[ryscheng]

Going to close and consolidate under https://github.com/opensource-observer/oso/issues/4160