AI Automation Engineer
Ibraheem
El Nahta
I build AI agents and automations that replace manual work, run them in production, and measure whether they actually work.
Everything below is in production at a boatbuilder in Bahrain, where I’m the only AI engineer: I decide what to build with management, then design, ship and run it.
- Agent accuracy
- 74% → 89%
- Assistant tools
- 50+
- University of Southampton
- MEng
- Mechanical Engineering / Mechatronics
- Languages
- EN · AR · MS
Selected work
Systems in production
Vera: a WhatsApp AI sales agent, and how I know it works
A bilingual agent that answers customers, qualifies leads and updates the CRM, measured by my own evaluation harness.
74% → 89% accuracy across two prompt revisions
- n8n
- Claude API
- Qdrant
- Chatwoot
- WhatsApp Cloud API
Read the write-up →
Command Centre: one portal for a whole company
Separate sales, marketing and operations tools brought into one portal, with an AI assistant and a management dashboard on top.
50+ tools the internal AI assistant can use
- React
- TypeScript
- Tailwind
- Supabase
- Vercel
Read the write-up →
Book Studio: writing an illustrated book as a team
A Next.js app where a team writes, illustrates and reviews a book together, with an AI assistant and a connector for people's own Claude.
1 editor per page at a time, with conflict-safe saves
- Next.js 16
- React 19
- Tailwind 4
- Supabase
- TipTap
Read the write-up →
V-CAPTAIN: an offline-first app for boat owners and crews
Maintenance, inspections, documents and warranty for many owners and boats, working offline at sea.
Offline works at sea, syncs on reconnect
- React
- TypeScript
- Supabase
- PWA
- GitHub Actions
Read the write-up →
A system that documents itself
An architecture map generated from the code on every push, with a compare view for reviewing a branch before it ships.
0 hours of manual upkeep
- Node.js
- TypeScript
- React Flow
- Supabase
Read the write-up →
Product launch sites that feed the CRM
Three server-rendered product sites whose enquiries go straight into the CRM, so no lead is lost.
3 product sites, one per model
- TanStack Start
- React
- Tailwind
- Zod
- Vercel
Read the write-up →
How I work
- I measure what I ship
- An AI agent gets a test suite and a score per version, not a few spot checks. If I can't show it works, it isn't done.
- Production standards from day one
- Audit logs, soft deletes with restore, automated checks on every push, and alerts so failures are loud rather than silent.
- Documentation another engineer can follow
- Plans and feature docs live with the code, and the system map is generated from the code itself, so it can't go stale.