Hey @nnnn thanks for the followup comment. There's a few considerations here:
- What types of queries are you running? From my experience + others, embedding-top-k retrieval helps to provide answers to fact-based queries but not others (summarization queries, queries where you want to synthesize across heterogenous data sources). In gpt index, list index is good for summarization, and people have also tried defining graphs over their data to enforce that certain information is fed in to generate results.
- as a general note, if you're building an app that depends on data, you're going to want nice data structure support, and as a python package gpt index can provide that. typescript's data structure support doesn't seem as good from what i can tell.
- we offer abstractions for response synthesis so that you don't have to worry about token limitations given retrieved context
- we offer abstractions around vector store support that will make it easy to to swap out different vector stores if you were trying to test them out
