So my guess would be you have both a inverted index and some sort of ANN index deployed, and route queries to those depending on the inferred intent?
Could be at least 1 order of magnitude faster though.
There are just the two of you working on typesense right? Surely having some more people working on the core product (up to five?) should help, no?
I'm also building an Algolia alternative, with three engineers including me working on the core engine, and sometimes I wish I had one or two more.
This really resonates with me, I'm currently bootstrapping an Algolia alternative too, albeit with the Saas business model instead of open core. All the best!
Meilisearch is rather the free and open alternative to Algolia, as it focuses on solving zero-config and easy to deploy autocomplete search, as opposed to analytics at scale.
Are you using Opensearch for logging or site search? $371 seems like a lot for a low traffic site. You can probably reduce the cost to a fraction by doing some research on alternatives
Yes doclist compression is a must. Higher intersection throughput and less bandwidth stress. Are you loading your doclists from persistent storage? What is your current max rps?
I assume you are using Symspell for fuzzy search and prioritize candidate frequency over edit distance. My input "algolai" returns results for "angola" even though "algolia" was just 1 transposition away. Anyway just dropping in to say that stuff on your github helped me figure out a lot while building a search engine from scratch. Good luck with SeekStorm!