LlamaIndex can't do chunk-level metadata, only document-level metadata, so you can't put precise references to where materials the LLM synthesized answers from originated, e.g. HTML anchors. Just write your own RAG with Pinecone and OpenAI APIs directly.