@iarafah3
Founder & CEO, CordaPoint

"Vibe coding" works great for weekend demos.

It falls apart in production—especially when you're building data infrastructure for Arabic AI.
AI-assisted coding lets you ship features in minutes, but an interface or annotation tool can look completely finished in English—clean layout, passing types—and still break the moment real right-to-left Arabic meets numbers, code-switching, or dialect edge cases.

And if a vendor's data pipeline is sloppy under the hood, your model pays for it.
At Cordapoint, I'm lucky to build alongside a sharp remote engineering team using Claude Code—and we hold every single line to the same bar: scoped branches, pull requests, peer code review, and automated checks before anything touches production.

What that discipline built under the hood:
🔹 RTL-native architecture: TypeScript, React, Tailwind, and Radix UI engineered for true bidirectional English/Arabic parity—not patched on as an afterthought.

🔹 Strict schema boundaries: Runtime validation with Zod across Cloudflare Pages, Functions, D1, and KV so every contributor step, dialect tag, and review decision is structured and traceable.

🔹 End-to-end vetting & delivery: A recorded interview system for native dialect speakers paired with a client portal built for transparent batch delivery.

I put the full stack and architecture breakdown on my GitHub profile—link in the first comment.
If you're training Arabic LLMs, speech models, or evaluation benchmarks and want to collaborate on a pilot or custom dialect project—DM me or comment with what you're building.

#ArabicNLP #AIData #LLM #DataQuality #BuildInPublic

Member ReposAI Builders ToolkitSkillsModel Benchmarks

Comments1

  • @iarafah
    Arafah Al Khateeb
    Founder of CordaPoint — Native Arabic–English data, review & translation for AI teams. Building fair, recorded hiring tools for NGOs.
    36m ago

    Full breakdown of the stack and what we've built so far: github.com/iarafah

Log in to join the thread.