Classification
The product, jurisdiction, question type and required source groups are identified.
PROTOTYPE · LEGAL AI · HUMAN REVIEW
A domain-specific RAG assistant scenario for questions at the intersection of law, banking products and principles of partnership (Islamic) finance. Its role is to find and compare relevant fragments and then produce a verifiable draft for specialist review.
Domain problem
When analysing a product, it's important to distinguish the actual description of the transaction, the applicable law, contractual terms and Sharia principles. The language model must not mix these layers or issue a confident conclusion without grounding in the source corpus.
Verification workflow
The product, jurisdiction, question type and required source groups are identified.
Hybrid retrieval finds exact terms and semantically close provisions in a trusted corpus.
The answer separates found provisions, source differences and areas where data is insufficient.
The draft and source list are passed to a person for legal and domain assessment.
Useful scenarios
Murabaha, riba, Sharia committee and other terms with the found basis cited.
Transaction structure, party roles and items needing further review.
A summary of relevant provisions without hiding discrepancies and gaps.
Structured material for further work by a lawyer or domain expert.
Source, score and chunk ID to return to the original document.
The system must show when the corpus isn't enough for a substantiated answer.
Quality principles
Corpus is bounded — only documents included in the database are used.
Source is visible — found fragments accompany the answer.
Confidence doesn't mask gaps — missing basis must be flagged.
Roles are separated — the system searches and drafts; the specialist decides.
Status and boundaries
The public page describes the project approach and the validated RAG architecture. The assistant does not issue binding opinions, replace a lawyer, compliance, a Sharia board or a fatwa, and is not used for automatic approval of financial products.