technical-blog

Tag

Cards List
#technical-blog

Pruning RAG context down to what the answer actually needs

Hacker News Top · yesterday Cached

Kapa.ai describes how they added a small LLM pruner between retrieval and generation in their RAG pipeline, which drops 68% of context while maintaining 96% recall, cutting query costs by a third.

0 favorites 0 likes
#technical-blog

The (Petty) Reason We Didn't End Up Using jj

Lobsters Hottest · yesterday Cached

Gradle explains why they decided not to adopt Jujutsu (jj) as their primary version control system: jj does not honor .gitattributes, causing persistent phantom modifications on files like gradlew.bat on Windows due to line-ending mismatches. They continue using git worktree instead.

0 favorites 0 likes
← Back to home

Submit Feedback