$ cat case-studies/flowcraft.md · 04 / 04
FlowCraft
$ what it had to do
Thejob.
One prompt gives you one shot at good writing. FlowCraft breaks the job into stages, outline, draft, polish and more, and chains them so each stage builds on the last: 4 templates, 5 writing personas and 25 chained prompts, at about 500 tokens a second on Groq. Around it is a real SaaS: sign-in, payments, a credit system and session history.
$ trace --request
How itworks.
- 01Pickone of 4 templates and 5 personas
- 02Chainstages that feed each other
- 03Generateat ~500 tokens/s on Groq
- 04Keepevery session, paid for in credits
$ git log --grep=decision
The callsthat mattered.
- decision.1Stages, not one prompt
Each stage does one job and hands its output to the next, so the result is built up the way a writer would build it.
- decision.2A real product around it
Clerk for sign-in, PostgreSQL with Prisma for data, Stripe for payments and a credit system, so it runs as a business, not a demo.
- decision.3Fast model, fast loop
Groq’s speed keeps a multi-stage pipeline quick enough to use, at around 500 tokens a second.
built with
- Next.js 14
- TypeScript
- Groq API
- PostgreSQL
- Prisma
- Clerk
- Stripe