Skip to content

Instantly share code, notes, and snippets.

@ruvnet

ruvnet/gist479-1.md

Created June 17, 2026 22:05
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save ruvnet/f78beca458b506b9c6356f27ee8a75c2 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save ruvnet/f78beca458b506b9c6356f27ee8a75c2 to your computer and use it in GitHub Desktop.
Download ZIP
Productionizing ruvector gnn-rerank (#479) — Step 1: baseline
Raw
gist479-1.md

Productionizing ruvector gnn-rerank (#479) — Step 1: baseline

What: A 1-hop GNN score-diffusion reranker over approximate-ANN candidate sets that recovers recall lost to quantization. Measured in the source research (#479): recall@10 28.0% → 38.4% (+10.4pp), N=5K D=128, still ~millions of QPS.

Step 1 (implement/integrate):

  • Pulled crate ruvector-gnn-rerank into the ruvector workspace.
  • 4 reranker variants: NoisyScore (baseline), GnnDiffusion (the win), GnnMincut, ExactL2.
  • 14/14 unit tests pass.
  • Branch: feat/productionize-gnn-rerank.

Honest caveat: the standalone benchmark binary is blocked by endpoint security (CrowdStrike) on the dev box, so the recall table is carried from #479's CI rather than re-run locally. Next step turns that delta into a CI-runnable regression test so the win is guarded, not just documented.

Next: recall regression test → criterion bench → security hardening → optimization → publish.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment