MediaWiki Code2Code Search: Neural Retrieval for Semantic Discovery of Open-Source Software Entities

The paper presents MediaWiki Code2Code Search, a neural retrieval system for semantic code-to-code discovery across large open-source ecosystems. It indexes 1.29 million structural entities, including functions, types, and templates, from more than 2,500 MediaWiki repositories. Its split-build architecture separates GPU-heavy offline indexing from CPU-only serving. A FAISS IVF-PQ index occupies 168.6 MB, reportedly 96.6% smaller than a flat float32 baseline, while median query latency is 1.85 seconds on commodity hardware. On a 27-query benchmark, it reaches P@10 of 0.87 versus 0.64 for BM25.