1 of 22

Arabic LLMs at the Core of Workflows

Arabic.AI Techniques and Approach

2 of 22

Arabic LLMs at the Core of Workflows�Arabic.AI Techniques and Approach

3 of 22

Agenda

Introducing Arabic.AI�and Nour Al Hassan, CEO of Arabic.AI group

1

Evolution of Arabic LLMs and why they matter

2

Limitations of standalone LLMs

3

Reality

4

Automation

5

High-level Overview

6

Agent Orchestration Patterns

7

Observability & Evaluation

8

Key Takeaway

9

4 of 22

2008

Founded

Launched

2021

Launched

Expanded to�Qatar

Rebranded to

2024

Introduced

Arabic LLM

Introduced

Content & Design Hubs

Introduced

Staffing Solutions

Launched

Rebranded to

2017

2023

Launched

Acquired

Expanded to�KSA

Launched

2022

Launched

Launched

2025

2014

Acquired by

Nour Al Hassan

The journey from

to

5 of 22

The evolution of

&

Roadmap:

Services Business: The Launchpad

Leading Arabic language services provider with deep sector expertise and a broad range of offerings.

Trusted by 700+ clients over 16 years, with $20M revenue, 7 MENA offices, and recent US expansion.

Data Annotation & Labeling:

Comprehensive data annotation services for AI/ML—rapid, scalable, and quality-driven, supported by a large vetted freelancer network and adaptable workflows.

Quality assurance and bias mitigation guarantee robust datasets for internal and external AI/ML projects.

World-Class Tech Team: The Innovators

Expert team of 65+ engineers, AI/ML specialists, and linguists with strong Arabic language and cultural expertise.

Drives innovation, customization, and ensures top-tier quality, security, and compliance in all services.

LLM Development & Fine-Tuning

Flagship Arabic LLM: leads in Arabic performance, dialect coverage, and specialized domains.

Customization & deployment: Fine-tunes models for enterprise needs using proprietary data, enabling critical sector applications.

Agentic Platform: The Future of Automation

First agentic AI platform in the region—Arabic.AI enables autonomous, multi-step workflows entirely in Arabic.

Automates complex enterprise processes—including document summarization, insight extraction, and customer support.

Seamless integration—scalable and adaptable deployment with client systems.

Growth & Vision

7 regional offices and recent US expansion, positioning us as a global Arabic AI leader.

1

2

3

4

5

6

6 of 22

Introducing

Ureed.com

Our freelancer Marketplace to allow for resource scalability. Connected with CleverSo TMS to seamlessly find, assign and pay translators digitally.

(Also known as: CleverSo)

First Arabic-centric Translation management system (TMS) geared towards the delivery of AI- enabled translation services

(Also known as: T-Portal)

The Gateway for our clients to submit language service orders and content digitally, while also tracking their orders and spending.

Arabic.AI Core

  • Arabic.AI LLM X
  • Agentic AI Platform

(Also known as: T-Staff)

Digital management of client service orders. Seamlessly connected to our CRM and finance system to allow for automatic data sync of client information and automatic invoice generation.

  • Applications

7 of 22

Who We Are

Arabic-First AI Platform

Purpose-built for Arabic language understanding with native support for dialects and cultural context

Industry Focused Solutions

Specialized in government, finance, retail, and professional services sectors across the region

Enterprise-Grade Security

ISO 27001 certified with GDPR compliance and on-premise deployment options

16+

Years of Arabic�Language Data

40K+

Expert-Vetted�Content

95%

Arabic�Accuracy

3X

Faster Than�Global LLMs

Powered by

Arabic.AI LLM (Pronoia) is the world’s #1-ranked Arabic AI model, our technology is purpose-built for the Arabic language. By combining cutting-edge AI with human expertise, we create scalable solutions that drive efficiency, innovation, and growth.

Why

Accuracy in Arabic

Proprietary Pronoia LLM + Arabic-native retrieval with citations.

Governed at Scale

SSO/SAML, RBAC, audit logs, data residency options (incl. KSA).

Time-to-Value:

No-code Agentic Platform builders, templates, and managed rollout.

8 of 22

Why Did We Pivot Into AI

So… it was a natural progression of products and services spanning over 16 years.

See a Need, Fill a Need!

If there is no one filling a need, then we must do it. Not because of money, because we want to make an impact.

Strong Pride in Our Heritage

Like the great poets of our past, we take pride in our service to our language.

Belief in Local Talent

We have been fortunate to have very talented people who were working on AI even before it became a trend.

9 of 22

New Products Evolution

Application

Arabic.AI | Customs | Social analysis

Platform

Workflows | Agents | Orchestration

Models

Speech | Translation | LLM | Vision

10 of 22

Business Model

Revenue Streams

  • Platform licensing + usage-based compute fees.
  • bundled professional services for specific use case deployments (Gov AI transformation, BFSI intelligent automation, etc.).

Customer Value Proposition

  • Packaged vertical solutions (use case + platform + managed services).
  • Deployed on-premise or private cloud.

Deal Structure

  • Multi-year contracts.
  • Sold via direct enterprise/government sales and SI.

11 of 22

Limitations�Of Standalone LLMs

Stateless interactions

Each request is handled independently, without guaranteed memory of previous interactions—unless context is explicitly provided again.

Broken workflows

Multi-step tasks (e.g., approvals, long reasoning chains) require external state handling.

No goal persistence

  • Drift from the original objective
  • Optimize for the latest prompt, not the overall mission
  • Contradict earlier decisions or plans

12 of 22

Reality Is Broken!

Scale is the game, then HS wins

Giant companies are training general purpose gigantic models

Most real-world problems are domain-specific, not global

Capital, sets ceiling

Winning requires GPUs and infrastructure, which will lock out smaller teams and regions

Smaller teams agility

Leaderboards

Big models excel in general leaderboards

13 of 22

Automations Will Fix It�(Automate everything)

DevOps Automation

Automate building, deployment and testing

Cookie Cutter

Vibe Coding with Templates: From CookieCutter to Production

Automate Workflow Evaluation

Monitor and validate pipeline performance automatically

14 of 22

Evaluation Demo

15 of 22

High-Level Overview

16 of 22

Software Engineering Tips

Software engineering

The application of an empirical, scientific approach to finding efficient, economic solutions to practical problems in software.

Optimization and trade-offs

  • Game of optimization and trade-offs
  • Replace bad ideas with better ones
  • Improve good ideas

Soak test

  • Load test
  • Run longer
  • Important to measure scalability and reliability

Learning

You must become an expert at learning and expert at managing complexity.

Complexity is needed!

Right product -> Product right

Value

Two-thirds of ideas produce zero or negative value. So Fail fast

17 of 22

Complex systems

Fruits of working through problems. Exploring solutions, But don’t start with over-engineering

Tech stack

  • All works, opensource, modern ones win
  • Supported by community
  • Less effort when vibe (Because the training dataset is bigger)
  • Design patterns matter
  • Big balls of mud

Decoupling

More code

DevOps tips

Virtualization VS Containers

More code

CI/CD

Decoupling

Show them our cluster, delete pods, …

Microservices

DORA

Software Engineering Tips

18 of 22

AI Terminologies

RAG

Embedder

RL

ReAct

Agentic AI Platform

Frontier VS Opensource

19 of 22

Go Over Them

Live Demo

20 of 22

Agent Orchestration Patterns

Baseline Pattern

1

Hierarchical

2

Planner & Executor

3

Peer to Peer

4

Role Based

5

HITL

6

21 of 22

Key Takeaways

Arabic LLMs Are The Engine

1

Autonomous Workflows Are The Vehicle

2

Software Engineering Is The Road

3

22 of 22

THANK

YOU

© 2026

arabic.ai

info@arabic.ai

+966 115207125