Thanks Mark for talking about hex2txt!

Re: context lengths, I just released a Livebook (“hex2context”; announcement tweet) that uses RAG to include only the most relevant doc snippets for the task-at-hand. It’s very basic right now but in my (limited) testing it does add value. All embeddings are computed, stored, and queried locally, but I’d be interested in evals comparing different models (including paid ones).

Direct link to run the notebook: Run - Livebook.dev

(Also linked now from the https://hex2txt.fly.dev homepage.)

@mjrusso,

Great use of RAG! Love it! Thanks for sharing what you’re working on.

@mjrusso you may be interested in this PR, just a proof of concept at the moment and needs some more work, would love to have your input on it or even collaborate on the code: https://github.com/elixir-lang/ex_doc/pull/1976