Plate I · Legal & regulatory research
Hukum-AI
Legal and regulatory research
Arabic legal research where every statement resolves to its source.
In one paragraph
Hukum-AI is Siyada Tech's Arabic-native legal and regulatory research system. It answers questions over Saudi laws, regulations and internal policy, returns every statement with a link to the passage it came from, and says so when the sources do not support an answer. It runs inside the customer's environment with a full audit trail.
01What it is
A retrieval-augmented research assistant for Arabic and English legal and regulatory material. A user asks in Arabic or English; Hukum-AI retrieves the relevant passages from the connected corpus, drafts an answer, and attaches the citation behind each statement. What it cannot ground in a source, it returns unanswered rather than guessed.
02Who it is for
- In-house legal teams at Saudi companies
- Compliance and regulatory affairs functions
- Legal and policy departments in government entities
- Law firms handling Saudi regulatory work
03The problem
Saudi legal and regulatory material is Arabic first, spread across many issuing authorities, and revised often. General chat models handle formal Arabic legal language poorly, cannot see internal policy, and produce fluent answers with no verifiable source, which is useless for legal work.
04How it works
-
01
Ingest
Laws, regulations, circulars and internal policy are parsed, segmented and indexed with Arabic-aware chunking.
-
02
Retrieve
Hybrid keyword and semantic search over the Arabic and English index, tuned for legal terminology and diacritic variance.
-
03
Ground
The model may use only the retrieved passages; every claim is tied to a passage identifier.
-
04
Cite
The answer renders with inline citations that open the exact source text.
-
05
Log
Question, retrieved context, model version and answer are written to the audit log.
05Where it runs
- In the customer's cloud subscription or data centre
- Private network and on-premises options where data cannot leave the environment
- Web interface and API, with sign-in through the customer's identity provider
- Document sources connected read-only from SharePoint, file shares or a document management system
06Security and data
- Personal data handling designed to PDPL and NDMO practice
- The corpus stays wherever the customer deploys; nothing is copied out
- Role-based access; retrieval respects the source system's permissions
- Immutable audit log of questions, sources, model version and answers
- Independent certification status: [EVIDENCE REQUIRED]
07Evidence
What we measure here. Results are published once each one has a stated method, sample size and evaluation date. How we publish evidence
- Citation accuracy on Saudi regulatory queries
- Answer groundedness rate
- Research time per matter
- Corpus coverage (documents and issuing authorities indexed)
08What it is not
- It is not legal advice, and it does not replace a qualified practitioner's judgement.
- It does not know documents it has not been given; answer quality is bounded by the corpus.
- Recent regulatory changes appear only after the corpus is refreshed.
- Scanned documents need OCR good enough for reliable Arabic extraction.
- We do not claim zero hallucination. Grounding and citation reduce unsupported statements; measured rates are not yet published.
?Asked often
Questions
Does Hukum-AI work in Arabic?
It is Arabic-native: retrieval, segmentation and answer generation are built for Arabic legal language, with English in the same index.
Where is the data stored?
Inside your environment. The corpus and the audit log stay under your control and your residency rules.
Does it cite its sources?
Every statement carries a citation that opens the exact passage used. If no supporting passage is found, it says it cannot answer.
Is it PDPL compliant?
It is designed to PDPL and NDMO practice, and deployment keeps personal data in your environment. Compliance is assessed for each deployment with your data protection officer.
Does it replace lawyers?
No. It shortens research, not judgement. Its output is written for a qualified reviewer.
How do we start?
A scoping call, then a proof on a defined corpus and a real set of questions from your team.
See it on your own corpus
We run a scoped proof on your documents and your real questions, then walk you through the citation trail.
Last reviewed · Siyada Tech engineering