We replaced fixed-line chunking with syntax-aware boundaries for Go and JavaScript/TypeScript. This approach align chunks to top-level declarations and associated documentation instead of arbitrary byte counts, significantly improving context relevance for AI retrieval.