Good breakdown of the two paths. The part that trips people up in production is not the diagram itself, it is keeping the indexing path in sync when source documents change after the index is built. That drift causes more retrieval failures than bad chunking does.
Yes, exactly! That drift is one of the most overlooked causes of retrieval failures. I plan to cover that too, sometime later in the series. Thanks for bringing that to attention here.
Looking forward to that piece, indexing drift is one of those things that only becomes obvious once something breaks downstream and nobody can explain why the answer changed.
Good breakdown of the two paths. The part that trips people up in production is not the diagram itself, it is keeping the indexing path in sync when source documents change after the index is built. That drift causes more retrieval failures than bad chunking does.
Yes, exactly! That drift is one of the most overlooked causes of retrieval failures. I plan to cover that too, sometime later in the series. Thanks for bringing that to attention here.
Looking forward to that piece, indexing drift is one of those things that only becomes obvious once something breaks downstream and nobody can explain why the answer changed.
This is really well written
I'm not a technical person, just a laperson, and I was able to follow it very well. Thank you. I love understanding what's under the hood.
Thanks for your reply! Very glad to know that you found the post useful :)