Founders

sharing the % breakdown of the cost to build on top of Farcaster:

1. fast database (30%)
2. an indexer for hub (20%)
3. AI (15%)
4. recommendations (5%)
5. data inference, API, etc. (~10%)

majority of early cost will be running a performant database ecosystem.

It looks like a bad architecture to me. Is the Indexer something like elastic? What is the number of transactions processed? 
It could be actually the other way around  (very solid reco engine and API layer) but I haven't seen anything like that. Are there any details, like techstack, db size, transaction counts and so on?

Seasoned software development leader. OG. Acc. 
https://adamsobotka.deno.dev/

This looks like a very sane setup. What's the PG autoscaler, something from Google? Anyway, I'll be happy to jump on a short call to discuss this, I know very little about GCP, but the rest is kinda standard SaaS architecture (except shuttle off course) and I can't see the reason why the db should eat that much.

indexer is Farcaster shuttle, processes about a few hundred thousands transactions a day.

the recommendation and API components have no load so not much % cost going to them.

GCP, Postgres, PubSub, NestJS, serverless wherever possible