Geek Out Time: The “Rich” Get Smarter in RAG-Continuing from “Embeddings”

Geek Out Time: The “Rich” Get Smarter in RAG -Continuing from “ Embeddings ”

In the earlier two blogs : 𝗚𝗲𝗲𝗸 𝗢𝘂𝘁 𝗧𝗶𝗺𝗲: Chunk Happens — Testing different Chunking Strategies for RAG ( https://medium.com/the-constellar-digital-technology-blog/chunk-happens-testing-different-chunking-strategies-for-rag-5f15e804c8de ) and 𝗚𝗲𝗲𝗸 𝗢𝘂𝘁 𝗧𝗶𝗺𝗲: Embeddings -Continuing from “Chunk Happens” (RAG) ( https://nedvedyang.medium.com/embeddings-continuing-from-chunk-happens-rag-c7ad75925e37 ) , we looked at how different chunking strategies shaped retrieval in a RAG pipeline. That work showed something important: where you cut the text matters a lot. But I left with a nagging thought — is chunking the full story?

Embeddings are only as useful as the information they encode. If we embed bare chunks, the vectors reflect shallow slices of meaning. What if we enriched each chunk with summaries, keywords, and causal links before embedding? Would the retrieval system behave differently? Would it actually feel smarter?

Setup

This is an excerpt — the full article continues on Medium.