Arabic AI
Arabic AI assistants that answer from your documents and show the source
Seekers builds bilingual Arabic and English knowledge assistants (Arabic RAG) and AI agents that read your policies, contracts, and correspondence, then answer with a citation to the exact page. They run privately in your environment, and we test them on your own Arabic questions before anyone relies on them.
Definition
What is Arabic RAG?
RAG stands for retrieval-augmented generation. When someone asks a question, the system first searches your documents for the relevant passages. A language model then writes the answer from those passages only and cites them. Arabic RAG is the same approach built to handle Arabic properly: spelling variants, word forms, scanned documents, and questions that mix Arabic and English.
Why do general AI tools struggle with Arabic documents?
Search built for English assumes each word has one spelling and a clear boundary.
Spelling variants
أحمد · احمد · إحمد
Mixed languages
تم approve الـ invoice
Franco-Arabic
3ayez a3raf el mo3ad
Dialects
أبغى · عايز · بدي
Scanned PDFs
OCR on stamped pages
RTL tables
Columns read right to left
Read moreShow less
Arabic breaks both assumptions. Prefixes and suffixes attach to the word, so وبالعقد ("and in the contract") and العقود ("the contracts") share a root but often fail to match in a keyword search. Writers also swap أ, إ and ا freely, along with ة and ه, and ى and ي.
Real enterprise content adds more problems on top:
- Mixed language in one sentence. A staff member types "ما هو الـ SLA في عقد الصيانة؟" and expects the system to find an English clause inside an Arabic contract.
- Scanned files. Many archives are scanned PDFs. Arabic OCR has to cope with connected letters, dots, and stamps before any of it can be searched.
- Right-to-left tables. A careless extractor reads an Arabic table in the wrong direction, and a fee ends up next to the wrong service.
- Dialects in customer messages. Messages arrive in Gulf or Egyptian dialect, or in Franco-Arabic (Arabizi) typed in Latin letters and numbers, such as "abgha a3rf mo3ad el tajdeed".
- Source pages. In regulated work, an answer that does not show the exact page it came from cannot be checked, so nobody can act on it.
How does Arabic RAG work, step by step?
Each step runs inside your environment, as part of the Understand layer of Seekers Core.
- Scanned PDFعقد التوريد
- Read (OCR)Text and tables extracted
- Normaliseأ، إ، آ become ا
- SearchFinds clause 7.2
- Answer + sourceCited · page 14
Read moreShow less
- Ingest: we connect to your file shares, document management system, and email archives, extract the text, run Arabic OCR on scanned pages, and keep table structure and page numbers.
- Normalize: we standardize letter variants and diacritics for search, and keep the original text for display and citation.
- Index: we split documents into passages and index them two ways, by meaning and by Arabic-aware keywords. Meaning catches paraphrased questions; keywords catch exact terms such as contract numbers.
- Retrieve: for each question, the system pulls the most relevant passages in either language, filtered by what that user is allowed to see.
- Answer: a private language model writes the answer from those passages only, in Arabic or English, with citations attached.
- Log: every question, retrieved passage, and answer is logged for audit and for later accuracy reviews.
How do answers cite their sources?
Each answer links to the document, page, and passage it used.
ما هي غرامة التأخير في عقد التوريد؟
0.5% of the order value per day of delay, capped at 10%.
Supply contract.pdf · Clause 7.2 · page 14
«تُفرض غرامة تأخير قدرها 0.5% عن كل يوم…»
Every answer shows its source
Read moreShow less
The user can open the source, see the highlighted text, and check it before acting. When the retrieved passages do not contain an answer, the assistant says so and lists the closest documents. It does not fill the gap with a guess.
How is accuracy measured before go-live?
We build a test set with your team before launch.
- Real questionsWritten with your team
- Run the systemSame questions, every release
- Score each answerCorrect, cited, safe
- Go / no-goAgreed threshold
Read moreShow less
It holds real questions that staff and customers ask, in Arabic, English, and both mixed together, each with the correct answer and the page it comes from. We add hard cases on purpose: dialect, typos, poor scans, and questions the documents cannot answer.
We run the system against that set and check four things: did it retrieve the right passage, is the answer correct, does the citation point to the right page, and did it decline when it should have. Your team reviews the failures and agrees the go-live threshold with us. The same test set runs again after every model or document update, so a drop in quality shows up before users notice it.
Which Arabic dialects can it handle?
The assistant reads Modern Standard Arabic, Gulf and Egyptian dialects, Franco-Arabic, and messages that switch between Arabic and English.
Read moreShow less
It replies in professional Arabic or English, following your style guide. If your customers write in other dialects, such as Levantine or Maghrebi, we test those on your own messages during the pilot before relying on them.
What can Arabic assistants and agents do?
- 01
Policy and procedure assistant
Reads
- Document management system
- file shares
- intranet
- +1
Does
Answers staff questions about HR, finance, and operations policies in Arabic or English, citing the clause and page.
Governed by
Each user only gets answers from documents they are allowed to see.
Read moreShow less
- What it does
- Answers staff questions about HR, finance, and operations policies in Arabic or English, citing the clause and page.
- Systems it connects to
- Document management system, file shares, intranet, and the HR system for role-based content.
- Governance
- Each user only gets answers from documents they are allowed to see. Every question and answer is logged.
- 02
Contract search
Reads
- Contract repository
- document management system
- supplier records in the ERP
Does
Finds clauses, obligations, renewal dates, and penalties across Arabic, English, and bilingual contracts, including scanned ones.
Governed by
Access by department and contract classification.
Read moreShow less
- What it does
- Finds clauses, obligations, renewal dates, and penalties across Arabic, English, and bilingual contracts, including scanned ones.
- Systems it connects to
- Contract repository, document management system, and supplier records in the ERP.
- Governance
- Access by department and contract classification. Read-only: it never changes a source contract.
- 03
Customer-message agent
Reads
- CRM
- ticketing system
- messaging channels
- +1
Does
Reads incoming messages in dialect or Franco-Arabic, works out the request, drafts a reply in professional Arabic or English, and opens or updates the case.
Governed by
Staff approve replies on sensitive cases before they are sent.
Read moreShow less
- What it does
- Reads incoming messages in dialect or Franco-Arabic, works out the request, drafts a reply in professional Arabic or English, and opens or updates the case.
- Systems it connects to
- CRM, ticketing system, messaging channels, and the customer knowledge base.
- Governance
- Staff approve replies on sensitive cases before they are sent. Personal data is masked in logs.
- 04
Correspondence drafting
Reads
- Correspondence management system
- document templates
Does
Drafts formal Arabic letters and memos from a short brief or an incoming letter, using your templates, references, and tone.
Governed by
An authorized employee reviews and signs every draft.
Read moreShow less
- What it does
- Drafts formal Arabic letters and memos from a short brief or an incoming letter, using your templates, references, and tone.
- Systems it connects to
- Correspondence management system, document templates, and email.
- Governance
- An authorized employee reviews and signs every draft. Version history is kept.
Arabic AI: frequently asked questions
Can it answer in English from Arabic documents, and the other way around?
Yes. Retrieval works across both languages, so an English question can find an Arabic clause. The answer comes back in the language of the question, and the citation points to the original text.
Does it work on scanned PDFs?
Yes, after OCR. Clean scans work well. Faxes, handwriting, and heavy stamps need more checking. During the Readiness Sprint we test OCR on a sample of your real archive and flag files that should be rescanned.
Will it make up answers?
Any language model can get things wrong. We reduce the risk by answering only from retrieved passages, requiring a citation, having the assistant say when it cannot find an answer, and measuring all of this on the test set before go-live. Users can check every answer against its source.
Where does our Arabic data go?
It stays in your environment. Documents, indexes, models, and logs remain on-premise, in your private cloud, or in a sovereign GCC cloud. The private AI page explains the deployment options.
How long does it take to launch an Arabic assistant?
A Readiness Sprint takes 2 to 3 weeks. A pilot for one use case takes 4 to 8 weeks, including building and running the test set.
Test an Arabic assistant on your own documents.
Start with a Readiness Sprint. We take a sample of your documents and real questions, test OCR and retrieval on them, and give you a plan for the pilot.