Gitmemory et al are awful. They've made me think seriously about the plausibility of building a custom search engine.
From what I've heard, the biggest problem with building a general-purpose search engine seems to be the index—indexing the entire web requires data centers many times larger than most can afford, and deciding what is relevant in such an index requires excessive training data. But while working, I typically only want my search engines to search the same 200-odd sites (GitHub issues, Stack Overflow, MDN, language docs, high-quality technical blogs, etc.) It's too many to know which "site:" query to use, but it's a small enough set that it should be a manageable problem for an upstart search engine.
It also seems like this is likely true in many domains. Taking on Google directly by trying to build a general purpose search engine seems to be a non-starter (most competitors just crib off of an existing player, and don't do much better). But a bunch of sharded search engines for specific domains could theoretically outperform Google/DDG/Bing within their respective domains, because the human using them has already narrowed the domain down dramatically. They wouldn't dethrone Google (general search is always going to be needed) but they could compete for a segment of traffic.