Use Case

Shariah-Compliant AI for Islamic Banks, Takaful, and Modaraba Operations

Pak-LLM supports Islamic finance teams by keeping Arabic and Urdu terminology consistent across contracts, flagging clauses that commonly breach Shariah constraints such as riba, gharar, and maysir, and producing bilingual summaries for a Shariah board to review. The model drafts and flags; a qualified Shariah advisor remains the decision-maker.

Written for:
Islamic banks, takaful operators, modarabas, Islamic windows of conventional banks, and halal certification bodies.
Last reviewed:

What does "Shariah-aware" mean for a language model?

It means the model carries a governance layer that checks outputs against explicit prohibitions — riba, gharar, maysir, and impermissible sector exposure — and cites the textual basis it used. Shariah-awareness is a review aid, not a fatwa: only a qualified scholar or Shariah board can rule on compliance.

  • Terminology is anchored to Arabic-Urdu banking vocabulary rather than translated ad hoc.
  • Outputs that touch a prohibition are intercepted and rewritten or escalated.
  • Citations point to Quran, Hadith, and classical fiqh references rather than being asserted without a source.
  • Uncertain rulings are surfaced as open questions for the Shariah board.

How does the Amanah Gate review work?

Amanah Gate is an interception layer that inspects model output before it reaches the user. When a response touches a prohibited concept it is blocked, rewritten with a compliant alternative, or escalated with an explanation of what triggered the flag.

  • Runs on every response, not as an optional post-processing step.
  • Logs the trigger category for audit without storing the customer's document.
  • Can be tuned per institution with an approved phrase and clause list.
  • Escalation output is written for a human reviewer, not for the end customer.

Can it keep Arabic and Urdu terminology consistent across documents?

Yes. Pak-LLM uses a custom Perso-Arabic tokenizer and an institution-specific glossary so that terms such as murabaha, ijarah, musharakah, and takaful are rendered the same way across contracts, product disclosures, and customer communications.

  • Glossary pinning per institution and per product line.
  • Bilingual side-by-side drafts for Shariah board review.
  • Consistent transliteration rules for Roman Urdu customer materials.
  • Terminology drift detection across a document set.

What are the limits I should plan around?

The system cannot issue a fatwa, cannot certify a product as Shariah-compliant, and cannot replace the Shariah board's governance. It also should not be the sole control for regulatory reporting to the State Bank of Pakistan.

  • No certification authority: compliance sign-off stays with the Shariah board.
  • Human review required for any customer-facing disclosure.
  • Regulatory returns are prepared, not auto-submitted.

Frequently asked questions

Is Pak-LLM Shariah certified?

Pak-LLM is a software platform, not a financial institution or a Shariah advisory body. Its Islamic finance features are reviewed against published Shariah governance principles, but product-level Shariah certification is issued by the institution's own Shariah board, not by Pak-LLM.

Can Pak-LLM replace our Shariah board?

No. It prepares drafts, flags prohibited concepts, and organises evidence for review. Rulings and sign-off remain with qualified scholars on your Shariah board.

Does it support Arabic as well as Urdu?

Yes. Arabic and Urdu are both supported, along with English, and the platform can produce bilingual drafts so Arabic-source terms and Urdu explanations stay aligned.

Where does customer data go?

Under the default configuration, inference is ephemeral and documents are not used to train foundation models. Banks needing full isolation can deploy the on-premise appliance so no document leaves their network.

Continue reading