ICP● Flagship
Architect & Tech Lead
Enterprise document-intelligence platform running on-premises for major regional clients. I led it from scratch to production.
Hussein Turfah is an AI Automation & Full-Stack Engineer based in Beirut, Lebanon, who builds production-grade AI systems — VLM/LLM pipelines, RAG, and intelligent document processing — that run reliably in enterprise production, not just in demos.
AI Automation & Full-Stack Engineer · Beirut, Lebanon
Most AI dies in the demo.
Mine ships.
I build the unglamorous part: the stretch after the proof-of-concept, when a system has to run every day without anyone watching it. Most of my work lives on-premises inside enterprises; quietly replacing manual work at scale, staying accurate on day 200, not just launch day. I care less about what a model does in a notebook and more about what it does in production, under load, when it matters.
Production results
In production, Hussein Turfah's document-intelligence work runs at 97%+ extraction accuracy across 5,000+ documents per day, cutting manual processing headcount by 75% and end-to-end processing time by more than 60%.
Extraction accuracy
ICP pipeline
Docs / day
Throughput in prod
Headcount reduction
Manual processing eliminated
Faster end-to-end
Processing time cut
Selected Work
Things I’ve shipped that are still running.
BeachBeds
Full-Stack Lead
A multi-region beach reservation platform; React Native apps, a Laravel API, and a conversational AI agent that books seats by chat. Live beyond Lebanon.
Beehive
Security & Full-Stack
Secure messaging for an NGO. I owned the architecture and built the end-to-end encryption layer myself.
Also built: a loan-management system with deep ERP integration, and FarmDuino, an IoT greenhouse prototype.
Flagship · Architect & Tech Lead
Inside ICP: intelligent content processing at scale.
An enterprise document-intelligence platform I architected and led from scratch, now in production for major regional clients. It turns a manual, headcount-heavy document workflow into an automated system that classifies, extracts, and routes at scale; and stays reliable under real production load.
My role
I owned the technical direction end to end: architecture, model strategy, the enterprise integration layer, and the team that shipped it. My focus was making it accurate enough to trust and stable enough to run unattended.
Outcomes
Extraction accuracy
97%+
Documents / day
5,000+
Manual headcount
−75%
End-to-end time
−60%
Deployed on-premises for enterprise clients. No public interface; this is internal production software, not a demo.
How I got here.
I lead engineering at Kipco (Kuwait), where I've taken three enterprise systems from whiteboard to production and run the team that keeps them there. Alongside it I lead full-stack for BeachBeds out of Beirut. Before that, two years building SaaS products at Dojo in Abu Dhabi; Next.js, real-time, security-hardened. The through-line: I don't hand off at the demo. I own the part where it has to work.
Résumé (PDF) ↓What I reach for.
For AI, I work in VLM and LLM pipelines, RAG, and the retrieval and extraction that makes them accurate enough to trust.
AI/ML
VLM · LLM · RAG · extraction · classification
Build
TypeScript · Python · React · Next · Node · Laravel · FastAPI
Enterprise
ERP · DMS · CRM integration · Docker · CI/CD