Building AlMustashar: The Struggles Behind Saudi Arabia's Legal AI
How we overcame the challenges of building an AI legal assistant for all Saudi law — from Arabic NLP hurdles to earning user trust in a high-stakes domain.
Sa7ab Team
Sa7ab Technology Solutions
When we set out to build AlMustashar, we knew the destination: an AI assistant that could make Saudi law accessible to everyone. What we didn't fully anticipate was the road to get there.
This isn't a polished success story. It's the real account of what it takes to build an AI product in a high-stakes domain, for a language that most AI companies overlook, in a market where trust is everything.
“Legal consultation in Saudi Arabia costs 200-500 SAR per session. We believed AI could make this 200x more affordable — but building it right was harder than we imagined.”
The Struggles
Making AI Understand Arabic Legal Language
Arabic is one of the most complex languages for AI. Legal Arabic adds another layer — formal, archaic phrasing mixed with modern terminology, all in a right-to-left script that most AI models treat as an afterthought.
Covering the Full Breadth of Saudi Law
Saudi law isn't a single code — it spans labor law, commercial regulations, personal status law, criminal law, real estate, and dozens of royal decrees and ministerial resolutions. Building a system that handles all of these accurately was a massive undertaking.
Accuracy in a Zero-Tolerance Domain
In legal consultation, a wrong answer isn't just unhelpful — it can be harmful. We couldn't ship a product that hallucinated legal articles or cited non-existent regulations. Every response needed to be traceable back to real law.
Earning Trust in a Conservative Market
Saudi users are rightfully skeptical of AI giving legal advice. We had to build not just a capable product, but one that communicates its limitations honestly and earns trust through transparency.
Designing an Arabic-First Experience
Most tech products treat Arabic support as a translation layer on top of an English product. We had to rethink the entire UX — from right-to-left layouts to Arabic typography — to make it feel native.
Making Legal Help Accessible, Not Just Available
Traditional legal consultation in Saudi Arabia costs 200-500 SAR per session and requires scheduling days in advance. We needed a pricing model that makes legal guidance genuinely accessible to everyone.
The Turning Point
The breakthrough came when we stopped trying to make a general AI smarter and started building a system purpose-built for Saudi law. Instead of fine-tuning a model to “know” the law, we built an architecture where the AI always consults the actual legal text before responding.
The Principle
“Don't trust the AI's memory. Make it look up the answer every single time — and show its work.”
This simple principle transformed everything. Hallucinations dropped dramatically. Users could verify every answer. And we could update the legal knowledge base without retraining the entire system.
What AlMustashar Solves
Today, AlMustashar serves as a comprehensive legal AI assistant covering all Saudi law — labor, commercial, personal status, criminal, real estate, and more.
Cited, Verifiable Answers
Every response includes specific article references (e.g., المادة 77) that users can verify independently, building trust through transparency.
Comprehensive Legal Coverage
From labor disputes to commercial contracts to family law — AlMustashar covers the full spectrum of Saudi legal domains in a single platform.
Document Analysis
Users can upload contracts and legal documents for AI-powered review, getting plain-language explanations of complex legal terms.
24/7 Instant Access
No appointments, no waiting rooms, no office hours. Legal guidance available anytime, from anywhere, on any device.
The Impact
What We Learned
Arabic AI is a massive, underserved opportunity
The Arabic-speaking world has 400+ million people and is woefully underserved by AI. Companies that invest in Arabic-first AI products now will define the market for decades.
Domain expertise beats model size
A smaller, well-architected system with strong retrieval outperforms a giant model that guesses. In high-stakes domains, architecture matters more than parameters.
Trust is built one accurate answer at a time
No amount of marketing replaces the moment a user verifies that every article you cited actually exists. Trust compounds — but one bad answer can reset it.
Accessibility is a business model, not a feature
Making legal guidance 200x cheaper isn't charity — it's unlocking a market of millions who could never afford traditional consultation. The demand was always there.
Building AlMustashar tested us in ways we didn't expect. But every struggle made the product better — more accurate, more trustworthy, more useful. We're proud of what it has become, and we're just getting started.