Edit this page | Blame

GN2 Integration with LLM search using cached results

Tags

  • assigned: jnduli, alexm, bmunyoki
  • keywords: llm, genenetwork2
  • type: enhancement
  • status: open

Description

We'd like to include LLM searches integrated into our GN searches, when someone attempts a Xapian search e.g. when I search for `wiki:rif group:mouse nicotine`, we'd do a corresponding search for `rif mouse nicotine` on LLMs, and show the results on the main page.

Another example:

xapian search: rif:glioma species:human group:gtex_v8 llm search: glioma human gtex_v8

This can be phased out into

  • [ ] 1. UI integration, where we modify the search page to include a dummy content box
  • [ ] 2. LLM search integration, where we perform a search and modify UI to show the results. This can either be async (i.e. the search results page waits for the LLM search results) or sync (i.e. we load the search results page after we've got the LLM results)
  • [x] 2.1 create a copy branch for the gnqa-api branch
  • [x] 2.2 create a PR containing all the branches
  • [ ] 2.3 how much would it take to get the qa_*** branch merged into main??
  • [ ] 3. Cache design and integration: we already have some

cache using redis (gn search history), so we may use this for the moment.

Let's use flag: `LLM_SEARCH_ENABLED` to enable/disable this feature during development to make sure we don't release this before it's ready.

Notes

The branch for merging to gn2:

https://github.com/genenetwork/genenetwork2/pull/863

The branch for merging to gn3:

https://github.com/genenetwork/genenetwork3/pull/188

(made with skribilo)