Hi everyone, I am currently building a semantic Q&A pipeline using Weaviate to help users troubleshoot installation errors and runtime scripts pulled from this resource. Everything seems to ingest properly at first, but I am running into a persistent issue where chunking mixed documentation that contains both natural language explanations and code snippets leads to significant embedding dilution. When querying specific error logs or version-dependent script behaviors, hybrid search either over-indexes on general setup paragraphs or completely misses exact keyword matches for error signatures. Along with that, we are seeing noticeable query latency spikes whenever we attempt to resolve cross-references between the parent article classes and individual snippet chunks under moderate read loads. Has anyone here found an effective chunking boundary or custom vectorizer configuration to preserve both exact technical tokens and overall semantic context for documentation hubs like this?