Nisarg GurjarEmail me
TESTFLIGHT · iOS

PromptCraft. It interviews you, then writes the prompt you meant.

Status
TestFlight — external testers on build 2
Type
Personal product
Stack
Expo · React Native · TypeScript · Vercel serverless · Claude
Scale
6 target-model profiles · streaming end to end
Year
2026
Streams live on TestFlightreal devices, real testers, real App Store pipeline
The problem

Most people don’t write bad prompts because they’re careless — they write them because articulating intent is hard. The fix isn’t a template library; it’s a short interview that extracts what you actually want, then writes the prompt for the specific model you’ll use.

What I shipped
  • Conversational intake — the app asks a few targeted questions, then produces a structured, model-optimized prompt between explicit markers, ready to copy.
  • Six target-model profiles — the same intent phrased for different model families, because they respond to different structures.
  • Live streaming — tokens render as they arrive, through a Vercel proxy that pipes the model’s stream straight to the phone.
  • History with swipe gestures, retry-with-context error recovery, and a shared-secret gate on the API so the proxy isn’t an open relay.
How it works
Question → interview → prompt
  1. 1You describe roughly what you want
  2. 2The app interviews you — a few sharp questions
  3. 3Claude streams the optimized prompt live to the screen
  4. 4Copy it out, or refine and go again
The hard part
Streaming on React Native’s new architecture. The standard fetch-with-ReadableStream approach is unreliable in production iOS builds — it works in dev and dies on device. The shipped fix streams over XHR with incremental progress events, which sounds oldschool and is exactly why it works everywhere. Thirteen UX refinements came from real tester feedback before build 2.
Proof
  • Live on TestFlight with external testers.
  • Streaming verified on physical devices, not just simulators.
  • Backend deployed and rate-gated on Vercel.
I need a prompt for summarizing legal docs
Who reads the summary — lawyer or client?
▍streaming your optimized prompt…
Where it stands

On TestFlight, iterating with tester feedback. The full Apple pipeline — bundle IDs, certificates, builds, review — done end to end.

I specified the app and the streaming proxy and took it through TestFlight review. Claude Code wrote both. How I work →

Want something like this built?

I take on projects where shipping something real — tested, verified, in production — is the whole point.

Email me