Architecture
AI MVP Architecture
Short answer
AI MVP architecture is a web app, a background worker, a model API, Postgres, and a meter. Not a mesh of agents and not a Python monolith you cannot hire for.
What is AI MVP architecture?
The minimum set of components that lets users log in, run an AI job, pay, and not share data with each other.
Components
| Component | v1 choice | Skip |
|---|---|---|
| App | Next.js | New JS framework |
| Data | Postgres + RLS | Graph DB |
| Jobs | Queue (or Vercel cron) | K8s jobs |
| Model | Hosted API | Self-host GPU |
| Obs | Sentry + usage table | Full APM suite |
Sync vs async
Chat can stream in-request. Embeddings, reports, and video must be jobs. If the tab close loses money, you needed a queue.
FAQ
Python or TypeScript?
TypeScript if the rest of the company is Next.js. Python only if the product is the model pipeline and you already have that team.
Related
Architecture review?
Send a diagram or a repo. We’ll mark what is v1 vs theatre.