Oryn - AI SaaS Platform
AI SaaS platform for multi-modal content generation
- Active users at launch
- 500+
- AI requests / day
- 1,000+
- API p95 latency
- ~180 ms
The Problem
Content teams needed a single workspace that could move fluidly between text, image, and code without context-switching across four different tools — and bill usage cleanly enough to charge for it.
Stack
- React 18
- Node.js
- OpenAI GPT-4
- DALL·E 3
- MongoDB Atlas
- Stripe
- JWT
- Vercel
What I Shipped
- Streaming GPT-4 responses with graceful retry/abort UX so slow generations don't block the editor.
- DALL·E 3 prompt studio with saved prompts, reroll, and resolution controls.
- Stripe subscription billing with metered overage on AI tokens.
- JWT auth with session refresh and a per-workspace role model.
Lessons Learned
Streaming responses are table stakes in 2025; without them, users perceive any >800 ms generation as 'broken' even when the final output is identical. Metered billing gets dangerous fast — cache aggressively and bill off the server tally, never the client event.