Explainable by Design

Every answer you can
show to your regulator.

Seif doesn't just give you the answer. It shows you the reasoning, the sources, the confidence, and the audit trail — because in a regulated environment, how you reached a conclusion matters as much as the conclusion itself.

The Problem

The Black-Box Paradox

Compliance officers face a specific paradox that generic AI tools were never designed to solve: the tools promise efficiency, but regulators require explainability. When a regulator asks how you reached a conclusion, "the AI said so" is not an acceptable answer — and it never will be.

Generic large language models produce answers with no audit trail, no citations, no confidence scoring, and no reasoning chain. They are trained to sound authoritative even when they are guessing. In most contexts that is fine. In compliance, a confident wrong answer is not just unhelpful — it is a regulatory breach waiting to happen.

The firms that deploy these tools are accumulating invisible liability. Every undocumented AI-assisted decision is a gap in your evidence file. Every answer without a source is an assertion you cannot defend. This is not a technology problem. It is a design problem — and it has a design solution.

"In a regulated environment, the right answer isn't enough. You need to explain how you got it."

The ComplianceAnswer Envelope

How Seif Explains Itself

Every answer Seif produces is wrapped in a structured envelope — five components that together make the answer verifiable, auditable, and defensible.

Answer Tier
DEFINITIVE · INTERPRETIVE · REQUIRES_REVIEW
What it means Classification based on source authority — primary law vs. guidance vs. uncertain terrain.
Why it matters You know immediately when to trust the answer and when to escalate to legal counsel.
Confidence Scoring
4 dimensions: authority, coverage, consistency, recency
What it means Not a single magic percentage — four independent scores that each measure a different facet of certainty.
Why it matters Transparent about where confidence is strong and where it is tentative. No false precision.
Reasoning Chain
Step-by-step audit trail
What it means What was retrieved, how it was filtered, what analysis was applied, and how the answer was synthesised.
Why it matters The full logic is visible. You can follow every step — and so can your regulator.
Citations
Exact text + document ref + paragraph + relevance type
What it means Every claim is anchored to the specific regulatory text it comes from.
Why it matters Verify any answer against the source material in seconds. Nothing is asserted without evidence.
Audit Hash
SHA-256, append-only, tamper-evident
What it means A cryptographic fingerprint of the answer at the moment it was generated.
Why it matters Prove to an examiner that an answer has not been altered after the fact.
Component Values What it means Why it matters
Answer Tier DEFINITIVE · INTERPRETIVE · REQUIRES_REVIEW Classification based on source authority — primary law vs. guidance vs. uncertain terrain. You know immediately when to trust the answer and when to escalate to legal counsel.
Confidence Scoring 4 dimensions: authority, coverage, consistency, recency Not a single magic percentage — four independent scores that each measure a different facet of certainty. Transparent about where confidence is strong and where it is tentative. No false precision.
Reasoning Chain Step-by-step audit trail What was retrieved, how it was filtered, what analysis was applied, and how the answer was synthesised. The full logic is visible. You can follow every step — and so can your regulator.
Citations Exact text + document ref + paragraph + relevance type Every claim is anchored to the specific regulatory text it comes from. Verify any answer against the source material in seconds. Nothing is asserted without evidence.
Audit Hash SHA-256, append-only, tamper-evident A cryptographic fingerprint of the answer at the moment it was generated. Prove to an examiner that an answer has not been altered after the fact.
The Architecture

Deterministic Where It Matters

Not everything needs a language model. When Seif can answer a question by traversing the regulatory knowledge graph — following typed relationships between obligations, permissions, risks, and mitigations — it does exactly that. Deterministic reasoning means the same question always produces the same answer.

LLMs are powerful, but their power is in synthesis and interpretation — turning structured knowledge into readable prose, drawing inferences across documents, identifying nuance. Seif uses them for that work. For facts the graph already knows, Seif uses the graph.

This hybrid approach is not a compromise. It is a deliberate choice: use the right tool for each part of the reasoning chain, and be transparent about which tool was used and why.

Graph Traversal

Obligations, permissions, rules — retrieved deterministically from 231K+ entities.

Semantic Retrieval

Guidance notes and circulars retrieved by meaning, not keyword matching.

LLM Synthesis

Structured knowledge woven into a clear, readable answer with full citations.

The Benchmark

Meeting the DIFC Standard

ADGM has not yet enacted a specific AI accountability framework, but the direction of travel is clear. The DIFC enacted Regulation 10 in September 2023 — the UAE's most detailed framework for AI systems processing personal data, establishing five pillars for accountable autonomous systems. Seif is designed to meet all five, so the firms using it are prepared regardless of which jurisdiction moves first.

Ethical
Regulation 10 requires
Unbiased algorithmic decisions
Seif's approach
Confidence scoring prevents false certainty. Answer tiers distinguish definitive answers from interpretive ones. The system tells you when it is uncertain rather than guessing.
Fairness
Regulation 10 requires
Equal treatment regardless of protected characteristics
Seif's approach
Firm-context personalisation is based on regulatory categories (firm type, activities, roles) — not demographic or behavioural data. All firms of the same type receive the same obligations.
Transparency
Regulation 10 requires
Non-technical explanations to data subjects
Seif's approach
Every answer includes a reasoning chain explained in plain language. Citations link to exact regulatory text. No hidden inference steps.
Security
Regulation 10 requires
Protection against data breaches
Seif's approach
Hash-chained audit trail. Append-only log. All data encrypted at rest and in transit. SOC 2 certification in progress.
Accountability
Regulation 10 requires
Internal governance and monitoring
Seif's approach
Full audit log exportable for regulatory examination. 7+ year retention. Every answer traceable to the query, retrieval method, sources used, and LLM calls made.
High-Risk Processing
Regulation 10 requires
Additional obligations for sensitive decisions
Seif's approach
The REQUIRES_REVIEW answer tier explicitly flags uncertain answers, routing them to human oversight rather than presenting them as definitive.
Reg 10 Pillar Requirement Seif's Approach
Ethical Unbiased algorithmic decisions Confidence scoring prevents false certainty. Answer tiers distinguish definitive answers from interpretive ones. The system tells you when it is uncertain rather than guessing.
Fairness Equal treatment regardless of protected characteristics Firm-context personalisation is based on regulatory categories (firm type, activities, roles) — not demographic or behavioural data. All firms of the same type receive the same obligations.
Transparency Non-technical explanations to data subjects Every answer includes a reasoning chain explained in plain language. Citations link to exact regulatory text. No hidden inference steps.
Security Protection against data breaches Hash-chained audit trail. Append-only log. All data encrypted at rest and in transit. SOC 2 certification in progress.
Accountability Internal governance and monitoring Full audit log exportable for regulatory examination. 7+ year retention. Every answer traceable to the query, retrieval method, sources used, and LLM calls made.
High-Risk Processing Additional obligations for sensitive decisions The REQUIRES_REVIEW answer tier explicitly flags uncertain answers, routing them to human oversight rather than presenting them as definitive.

See explainability in practice.

Book a 30-minute demo and we'll walk you through a live compliance question — from query to reasoning chain to audit hash.