The scenario
An employee asks whether a specific allowance applies to their grade under the latest circular. The correct answer depends on which version of the policy is in force and what their grade is.
Knowledge AI demo
Answers from your policies and contracts, cited, permission-aware, and evaluated
Governed retrieval over the documents your organisation actually runs on, with the clause shown beside every answer and two roles seeing two different results.
In Saudi Arabia
What turns the Knowledge AI demo into a production system
A working demo and a system a Saudi reviewer will sign off are two different things. These are the questions asked before go-live, and what we hand over against each one.
AuthorityWhat it requiresWhat we buildWhat you keep
SDAIA, PDPL
Lawful basis for every personal-data field the demo touches, before it runs on real records.
Field-level data map produced during scoping, sanitised extract used until that map is signed off.
Data map and lawful-basis note the client's DPO can file before production access is granted.
NDMO
Data classification and quality controls on the sources the demo would depend on in production.
Source-by-source classification, with quality thresholds agreed before the pilot widens.
Classification map and a quality gate report your data office keeps.
NCA ECC
Access control, logging and read/write boundaries clear before wider rollout.
Entitlements scoped per role, every action logged, write paths gated behind approval.
Control matrix and an action log your risk function can replay.
Where it runs
The demo itself can run on a sanitised or representative extract, but where the data cannot leave a system, the build happens inside your environment inside Saudi Arabia, not ours.
Arabic, as it is spoken here
Arabic questions and documents are handled as first-class input, not translated to English before processing, including organisational and Gulf-dialect terms.
Who it is for
Legal, HR, compliance and service teams whose answers live in circulars, manuals and contracts.
Sample workflow
The chain of work behind the answer
1
Ingest
- Policies, contracts, circulars, manuals
- Arabic and English, scanned or native
- Version and effective date captured
2
Structure
- Clause-level chunking, not blind splits
- Metadata: owner, scope, classification
- Superseded versions marked
3
Retrieve
- Hybrid keyword and semantic search
- Entitlement filter before ranking
- Arabic morphology handled properly
4
Answer
- Grounded response with the clause quoted
- Says “not covered” when it is not
- Links to the document and page
5
Evaluate
- Test set written by your experts
- Scored in Arabic and English
- Regressions caught before release
Retrieval quality is reported as a score on your own questions, so “it felt good in the demo” never becomes the acceptance criterion.
What you will see
Two examples, start to finish
“Does the housing allowance apply to grade 9 after the March circular?”
- Behind it
- Identity resolved → circular version in force selected → clause retrieved → grade condition evaluated → superseded text excluded.
- What comes back
- A direct yes or no, the clause quoted, the circular reference and effective date, and a note if an older version says otherwise.
“What is our termination notice obligation for this supplier?”
- Behind it
- Contract identified → termination clause and any amendment retrieved → conflicting annexes surfaced rather than merged silently.
- What comes back
- The notice period, the governing clause, the amendment that changed it, and a flag if two documents disagree.
Deliverables
What you keep after the session
Every item below is an artefact your team can reuse, whether or not the engagement continues.
Governed index of a real corpus
One department's documents, ingested with version and classification metadata.
Evaluation set and scores
Your experts' questions with graded answers, in both languages.
Permission model
How entitlements were mapped from your directory into retrieval.
Content gap list
Questions your documents cannot answer, usually the most actionable finding.
Session agenda
- The same question from two roles with different entitlements
- A question the corpus cannot answer, and how the system refuses
- An outdated version test: does it answer from the superseded text?
- Arabic and English asked side by side on the same clause
What we need from you
- A document set from one department, including scanned files
- Twenty questions with agreed correct answers
- Role definitions for the entitlement test
Where data cannot leave a system, the demo is built inside your environment instead of ours.
FAQ
Frequently asked questions
Does our content train a model?
No. Retrieval reads your content at query time inside your boundary; nothing is used for training.
How do you handle scanned Arabic?
Through the Document AI pipeline, Arabic OCR, table recovery and Hijri date handling, before anything reaches the index.
What if two documents contradict?
The answer shows both and names the governing one by effective date, instead of silently choosing.
Topic cluster
Explore the topic cluster
Curated links across our Saudi hubs, sector pages and long-form blog research.