Is it possible to have a VecterStoreIndex retriever over a set of indices, rather than an entire query engine? Basically I have 1 index per article, and I want to build a meta-index over them based on their article metadata and summaries, and be able to filter/retrieve top k article indices using vector search. ComposableGraph seems to only support as_query_engine() but not as_retriever(). My workaround would be to create a new Document (meta-doc) for each article index, containing their metadata only, and create an index+retriever over those and manually extract their index ids from the retrieved meta-docs...
this feels like it should be possible somewhere/somehow though 🤔 I know you can add indexes as nodes to indexes using the IndexNode. Maybe if you wrote a custom retriever that recursively retrieves?
I ended up just doing exactly what I described with creating a normal index over "metadocs", since I'm already keeping track of unique ids. But I briefly looked into the IndexNode you mentioned and it seems like that would be the way to go, I will fallback to that approach if my "workaround" is insufficient, thanks! :)
