Contextual Retrieval vs Traditional RAG – thoughts and experiences

[msbayindir]

Traditional RAG systems often lose critical context when documents are chunked.
Contextual Retrieval tries to fix this by attaching document-level and section-level context to each chunk.

I'd love to hear your thoughts:

• Have you experienced context loss issues in classic RAG setups?
• What techniques have you tried to mitigate it? (larger chunks, overlap, metadata, reranking, etc.)
• Do you think contextual retrieval is worth the extra cost/complexity?
• Any benchmarks, papers, or real-world experiences to share?

This discussion is meant to be open-ended and technical.
Different perspectives are very welcome 👇