Inspiration : Inspiration
Every content creator knows the pain. You spend hours — sometimes days — researching, writing, and refining a long-form blog post. You hit publish. And then you stare at six empty content fields:
A Twitter/X thread A LinkedIn post An Instagram caption A newsletter A YouTube script intro A blog summary
The content is already there. But adapting it for each platform feels like starting from scratch every single time. That frustration is what sparked ContentSplit. As someone building at the intersection of technology and content, I kept watching creators — and teams — burn hours on distribution work that should take minutes. The best ideas deserve to travel further, faster. We just needed a smarter way to move them. ContentSplit exists to solve exactly that. One article. Six formats. Zero rewriting.
What it does : ContentSplit is an AI-powered web app that repurposes long-form blog content into platform-ready formats — instantly.
A user pastes or uploads their blog post. ContentSplit's dual-AI pipeline processes the content and returns six ready-to-publish formats simultaneously: Output FormatPlatformThreadTwitter / XProfessional postLinkedInCaption + hashtagsInstagramNewsletter editionEmail / BeehiivScript introYouTubeShort summaryBlog / SEO The core product loop works like this: Long-form Input→ClaudeStructured Analysis→DeepSeekPlatform-Ready Outputs\text{Long-form Input} \xrightarrow{\text{Claude}} \text{Structured Analysis} \xrightarrow{\text{DeepSeek}} \text{Platform-Ready Outputs}Long-form InputClaudeStructured AnalysisDeepSeekPlatform-Ready Outputs Anthropic Claude API handles the intelligence layer — reading the post, extracting the core argument, key points, tone, and audience. DeepSeek API uses that structured analysis to generate each format with the right voice, length, and platform conventions. The result feels authored, not automated. Users can manage their content history, upgrade to a paid plan, and access all outputs from a clean, minimal dashboard. Payments are handled by Flutterwave, optimized for African markets.
How I built it :
ContentSplit was built as a full-stack AI web application using a modern, lean stack: LayerTechnologyBuild EnvironmentAntigravityFrontendReact + Next.jsBackend & AuthSupabase (PostgreSQL)HostingVercelAI — Intelligence LayerAnthropic Claude APIAI — Content GenerationDeepSeek APIPaymentsFlutterwave APILanguageJavaScript + CSS The build process was driven by Antigravity — an agentic development environment. Instead of writing code line by line, I wrote structured context files: a PRD, design token system, AGENT.md specs, and a section-by-section BUILD_GUIDE. Antigravity executed the build from those specs. The AI pipeline was the most critical engineering decision. Claude processes the raw input and returns a strict JSON analysis object — key themes, tone, audience, format suggestions. DeepSeek receives that object as typed context and generates all six outputs in parallel. Enforcing structured JSON between the two models was what made the outputs consistent and platform-accurate. Supabase handled authentication and per-user content storage. Vercel handled deployment. Flutterwave powered the freemium-to-paid subscription flow.
Challenges I ran into :
Prompt chaining without drift Getting two AI models to work in sequence without the second losing context from the first was harder than expected. Early outputs were generic because DeepSeek wasn't receiving enough structured signal from Claude's analysis pass. The fix was enforcing strict JSON output from Claude — giving DeepSeek clean, typed context rather than freeform text. Format fidelity per platform Each platform has invisible rules. A Twitter thread isn't just short sentences — it has hooks, numbering, and rhythm. A LinkedIn post has a specific emotional arc. An Instagram caption needs a CTA. Teaching the generation layer to respect these unwritten conventions consistently required extensive prompt iteration. Supabase auth + Next.js edge cases Handling authentication state across server-side rendering and client-side navigation produced edge cases around session persistence and protected routes. Resolved by centralizing auth state and implementing middleware-level route protection. Antigravity learning curve Antigravity's agentic workflow requires a fundamentally different mental model. The shift from writing code to writing context — and trusting the agent to execute — took real adjustment. The quality of the output is directly proportional to the quality of the spec you provide. Payment integration for freemium Designing the freemium-to-paid conversion flow with Flutterwave required careful thinking around webhook reliability, subscription state syncing with Supabase, and graceful handling of failed payments across different African mobile network operators.
Accomplishments that I'm proud of :
Shipping a working dual-AI pipeline. Getting Claude and DeepSeek to work in sequence — with structured handoff between them — and producing consistently high-quality, platform-specific outputs is something I'm genuinely proud of. It required product thinking as much as engineering. Building with agentic tooling end-to-end. ContentSplit was built almost entirely through structured specs fed into Antigravity. Writing a PRD, design tokens, and context files as the primary build artifacts — rather than writing code manually — was a new kind of workflow and it worked. Designing for African markets from day one. Flutterwave integration wasn't an afterthought. The payment layer, currency handling, and subscription flow were designed with African users in mind from the start — not retrofitted later. Going from idea to deployed product in a cohort timeline. Taking ContentSplit from a product concept through PRD, UI design, full-stack build, and live deployment within a capstone program timeline is an accomplishment the whole team is proud of.
What I learned :
Prompting is product design. How you instruct an AI model is as important as the interface around it. The prompt chain between Claude and DeepSeek went through dozens of iterations — each one teaching us something about how models interpret context, structure, and intent. Simplicity is a product decision. Content creators don't want to think about AI. They want the output. Every feature idea was filtered through one question: does this reduce friction or add it? Most ideas didn't make the cut. Documentation is infrastructure. Using Antigravity taught us that good specs aren't just for humans — they're instructions for your build agent. A well-written PRD or AGENT.md is as important as any line of code. African markets need African-first thinking. Integrating Flutterwave surfaced how different payment infrastructure is across regions. Currencies, mobile money flows, and trust signals require deliberate design — not assumptions borrowed from Western SaaS templates.
What's next for Contentsplit ai :
Direct platform publishing — push outputs straight to Buffer, LinkedIn, and Beehiiv without leaving the app Brand voice profiles — train ContentSplit on a creator's unique writing style for outputs that sound like them Team workspaces — collaborative content pipelines for agencies and content teams Analytics layer — track which formats drive the most engagement per piece of content Audio & video support — repurpose podcast transcripts and video scripts, not just blog posts
The mission stays the same: help creators do more with what they've already built.
"Your best ideas deserve to reach every platform — without the extra work."
Built With
- anthropic-claude-api-(ai-intelligence-layer)
- antigravity-(ai-powered-build-environment)
- deepseek-api-(ai-content-generation)
- flutterwave-api-(payments)
- javascript
- next.js
- react
- supabase-(auth-&-database)
- vercel-(hosting-&-deployment)