There’s a question I used to ask myself every time a new circular landed from the FSRA: Where exactly in my obligation register does this go?
I wasn’t asking rhetorically. I genuinely didn’t know. I’d read the circular, understand what it said, and then face the task of figuring out which of our existing controls mapped to the new requirement — and whether any gap existed that an examiner might flag.
The process was manual. It involved three tabs, two Word documents, a spreadsheet someone had maintained since 2019, and at least one phone call to a lawyer who charged by the minute.
That’s not a compliance programme. That’s archaeology.
The Problem No One Was Solving
I’ve spent ten years in financial services, working at firms regulated by the ADGM and FSRA. I’ve been the person who had to certify compliance programmes. I’ve sat through examinations. I’ve filed incident reports and Suspicious Activity Reports and Annual Compliance Reports.
In that time, I tried every RegTech tool that came to market. Most of them fell into one of two camps:
Camp 1: Too generic. These tools were built for firms regulated by the FCA or SEC. They had ADGM mappings bolted on — usually incomplete, often out of date. The obligation libraries didn’t reflect ADGM’s specific firm categories, its particular approach to CDD, or the FSRA’s inspection methodology.
Camp 2: Too manual. These were essentially document management systems with compliance branding. You could upload your policies. You could tag them against obligation categories. But every classification decision was yours to make. The software stored your choices; it didn’t help you make them.
Neither camp solved the actual problem: helping a compliance officer at an ADGM-regulated firm understand, precisely and in real time, what their obligations are and whether their current programme satisfies them.
What Made ADGM Different
ADGM has its own rulebooks. They’re substantial — thousands of pages across the FSRA’s General Rulebook, the Conduct of Business Rulebook, the Collective Investment Funds Rulebook, the AML Rulebook, and dozens more instruments, guidance papers, and circulars.
These aren’t summaries of FCA rules. They’re a distinct regulatory framework, developed by a regulator with its own approach to enforcement, its own firm category structure (Category 1 through 4B), and its own examination methodology.
A compliance officer at an ADGM fund manager doesn’t need help with FCA obligations. They need help with their obligations — the specific combination of rules that applies to their firm type, their regulated activities, their client base.
That intersection isn’t documented anywhere in one place. It has to be derived: from the rulebooks, from the FSRA’s guidance, from the firm’s licence conditions, from the applicable scope provisions in each instrument.
That derivation was always done by hand. Until now.
Why AI Made This Possible
I’d been thinking about this problem for years. What held me back wasn’t motivation — it was capability.
The task of reading 8,000 pages of rulebooks, extracting every obligation, mapping each one to its applicable firm types and regulated activities, cross-referencing the definitions, and keeping the whole structure current as rules changed — that’s a knowledge engineering problem. You couldn’t do it with keyword search. You couldn’t do it with a simple database.
What changed was the combination of large language models and graph databases. Together, they made it possible to:
-
Extract structured obligations from unstructured rulebook text — not just summaries, but deontic statements: what the rule requires, what it prohibits, what it permits, for whom, under what conditions.
-
Build a graph of relationships — between obligations, definitions, rulebooks, cross-references, amendments, and precedent guidance. When a term like “beneficial owner” is defined in one instrument and used in twelve others, the graph knows that.
-
Map obligations to firm context — given what I know about your firm (category, regulated activities, products, client types), I can traverse the graph to return the specific subset of rules that applies to you.
-
Answer questions with citations — not “the answer might be somewhere in COBS” but “the obligation is in COBS Rule 4.3.1(a), which requires X, and here is the specific text.”
None of those capabilities existed at usable quality three years ago. They exist now.
What Seif Actually Does
Seif is not an AI that reads regulations and gives you answers. That exists. It’s called an LLM, and it hallucinates.
Seif is a platform built on three layers:
The knowledge graph — a structured representation of ADGM’s regulatory framework. Every obligation is a node. Every definition is a node. Every cross-reference is an edge. Every rulebook update is versioned. The graph is the ground truth.
The inference layer — when you ask a question, Seif doesn’t send your query to a language model and hope. It uses the graph to identify the relevant obligations, retrieves the exact text, and uses the language model only for synthesis — turning structured information into a readable answer.
The explainability envelope — every answer Seif produces includes the specific rules it relied on, the confidence level, the reasoning chain, and the audit hash. You can hand that to an examiner. You can hand it to your auditors. You can hand it to your board.
No black boxes. No “trust me, I’m an AI.”
The Compliance Officer I Was Building For
I had a specific person in mind when I built Seif. She’s a Head of Compliance at a Category 3C asset manager in ADGM. She has a small team — herself and one analyst. She covers AML, conduct, fund operations, and ad hoc regulatory enquiries.
She doesn’t have time to read every FSRA circular in detail. She doesn’t have budget for a law firm on retainer for every regulatory question. She needs to be able to look at her obligation register and know, with confidence, that it’s complete and current.
That’s who Seif is for.
When a new circular lands, Seif surfaces the specific obligation lines it affects and maps the change to her existing programme. When she has a question about whether a specific activity triggers enhanced due diligence, she gets an answer with citations. When an examiner asks her to demonstrate her compliance framework, she can export a package with evidence.
It’s not magic. It’s the thing that should have existed ten years ago.
What’s Next
We’re at an early stage. Seif currently covers ADGM and FSRA regulation — the rulebooks, guidance, circulars, and enforcement notices. We’re expanding firm type coverage and adding the DIFC framework.
The core principle won’t change: every answer Seif provides is grounded in the actual regulatory text, with the reasoning chain made visible, so you can trust it and defend it.
If you’re a compliance officer at an ADGM-regulated firm and you want to see how this works for your specific firm type and activities, book a demo. We’ll show you the obligation mapping for your firm and walk through a live regulatory query.
This post reflects my personal experience and perspective. Seif is not a law firm and does not provide legal advice.