from Hacker News

What if GitHub had vector search?

by snikolaev on 9/18/24, 2:11 PM with 2 comments

• by donhardman on 9/27/24, 9:55 AM

  This is interesting — has anyone tried something like this with Elasticsearch or maybe Milvus/Meilisearch/Typesense? I know those are popular for vector searches, but I haven't seen anything specific for improving GitHub searches like this.

  I'm curious how this compares performance-wise, especially when it comes to large repositories with tons of issues and PRs. Also, how scalable is it? I feel like semantic search has a lot of potential here, but does anyone know if GitHub itself has plans to integrate something similar?

• by costco on 9/19/24, 3:46 PM

  This is cool but would be even cooler if it could do code similarity search. The cost would probably be prohibitively high though unless you only did like 10000 repos. https://github.com/microsoft/CodeBERT/tree/master/UniXcoder#...