Repurpose AI
The Problem
One piece of content has to be rewritten for every platform that matters, and each platform has its own length, structure and tone. Doing that by hand takes hours, so most people either skip platforms or post the same text everywhere and get ignored.
Why Existing Approaches Fall Short
- •Writing platform specific versions by hand is slow and easy to postpone.
- •Generic AI output ignores the rules of each platform, so it has to be rewritten anyway.
- •Most tools produce finished copy when what people actually want is a strong first draft they can edit.
What Was Built
- ✓Four input modes: a rough idea, a transcript, a YouTube link, or a problem the audience has.
- ✓A strategy step that identifies the core message, audience, tone, hook ideas and content pillars before anything is written.
- ✓Nine output formats with platform rules built in, covering LinkedIn posts, X threads, short video scripts, cold email drafts, newsletter sections and more.
- ✓Inline editing for every draft, plus the ability to regenerate a single format without losing the others.
- ✓Four supporting tools: a format converter, a hook generator, a bio builder and a content angle finder.
- ✓Saved sessions, usage limits per plan and a small usage dashboard.
- ✓Accounts with email or Google sign in, with each account isolated from the others.
How It Works
Add the source
Paste a transcript, a video link, an idea or a customer problem.
Review the strategy
The platform sets out the message, audience and angles before producing drafts.
Generate and edit
Choose the formats you want, then rewrite each one in place.
Export the set
Copy everything at once or download the drafts as Markdown.
Key Architectural Decisions
Strategy before generation
Producing an outline first makes every later draft more consistent than asking a model for output immediately, and it gives the user something useful even if they only want the analysis.
Every format keeps its own rules
Character limits, structure and tone conventions are part of each prompt rather than general guidance, which is what makes the output publishable.
Treat AI output as untrusted text
All generated content is sanitised before it is rendered, and prompts are guarded against injection attempts in the source material.
Server side AI calls with automatic recovery
The AI key stays on the server. The proxy discovers which models are available, honours rate limits, retries briefly on transient failures and then fails with a clear message instead of a silent hang.
Stack & Rationale
React with Vite
A fast single page app that is quick to build and cheap to host.
TypeScript
Safer refactoring and fewer runtime errors.
Supabase
Authentication, database and per user data isolation.
Groq running Llama 3.3 70B
Fast generation at a low cost per request.
TanStack Query
Predictable data loading and caching without custom state handling.
Stripe, wired but not enabled
Subscription billing prepared for launch.
Vercel serverless functions
Keeps the AI key and payment secrets on the server.
DOMPurify
Sanitises AI output before it is displayed.
Lessons & Outcomes
- •Platform specific prompts create most of the value. The model is rarely the limiting factor.
- •People want drafts they can edit rather than finished copy, so regeneration matters more than perfection.
- •Handling AI service failures is part of the product, not an afterthought. Retired models and rate limits are normal events.