Arabic-first AI products
Aadhib Nasser · AI Engineer and AI Consultant
Who this is for
- 01Saudi organisations whose users work primarily in Arabic
- 02Teams with an English product that needs to become genuinely Arabic-first
- 03Legal, regulatory and other high-stakes work where an answer has to cite its source
What the work covers
- 01Arabic-first product design: right-to-left layouts, typography and bilingual navigation
- 02Retrieval over Arabic sources, with citations carried as data
- 03Source-grounded answers, and a designed refusal when no source supports one
- 04Bilingual content models that keep Arabic and English in step
- 05Multi-tenant architecture with tenant isolation enforced in the database
How I approach it
No source, no claim
Retrieval gates generation rather than informing it. If the corpus holds nothing that supports an answer, the system says so, and that refusal is a designed path with its own output.
Structure matters more in Arabic
Legal and regulatory Arabic is precise and structurally specific. Chunking it as though it were prose loses the article and clause boundaries that give a provision its meaning.
Designed in, not translated in
Arabic is a starting constraint on layout, type and content, not a toggle added at the end. Retrofitting it costs more than designing for it.
Related work
Case studies and projects
Further reading
Notes on this subject
Questions
Asked before an engagement
Can general models like GPT or Claude handle Arabic well enough?
For drafting and conversation they are capable. For work that has to be right, such as legal or regulatory questions, the limit is less the model and more the Arabic corpus it can retrieve from and cite. That is where most of the engineering goes.
Do you build bilingual products or Arabic-only ones?
Bilingual and Arabic-first: the Arabic experience is designed as the primary one, and English is built alongside it rather than the other way round.
Start a conversation
I usually reply within a day
Send a few sentences about the problem: what the process is today, where the data lives, and what would change if it worked.