A vector-only search engine will fail for a lot of common use cases where the keywords do matter. I tried searching for `garbanzo bean stew` and got totally irrelevant bean recipes.

Yes, indeed. I just tried search "Apple", and apple.com is not on the first page.

Agree. For best results both lexical and vector search results should be fed into a reranker. Slow and expensive but high quality.

What if you build a graph engine then encode those edges into its own embedding space?

Nerdsnipe?