We built Qanooni to solve a specific problem. When AI tools answer legal questions without showing the source, lawyers face a binary choice: trust the answer without verification, or verify it manually and lose the time saving the tool was supposed to deliver. We thought both options were wrong, and we built a third option: an AI that shows its work every time.
Where the problem started
The problem became concrete for us when we were watching a junior associate at a Dubai-based firm use a leading AI legal research tool for the first time. The tool was impressively fast. The associate asked about the notice requirements for terminating a UAE mainland employment contract and received a clear, confident prose answer in about ten seconds. The answer looked right. The associate typed it into the research note and moved on.
The answer was wrong. Not completely wrong; the general framework was accurate. But the specific notice period cited for the associate's scenario was drawn from a provision that applied to a different employment contract duration. The applicable provision was different. The error was not obvious from the prose output because nothing in the response indicated which provision had been used, or how it had been applied to the specific facts.
That incident is not unusual. It is a structural feature of how most AI legal research tools are built. The tools are trained to produce helpful, fluent answers. They are not trained to surface the specific source text behind each claim. The result is an output that looks like verified research but is not, because verification requires a source, and the source was not provided.
The verification problem is not a UX preference
We want to be precise about what kind of problem this is. Some product teams in the legal AI space describe the absence of citations as a user experience choice: lawyers want clean prose answers, not cluttered footnotes. There is some truth in this for low-stakes orientation tasks. But the verification problem is not about user preferences for footnotes. It is about professional accountability.
A practitioner who gives advice to a client is professionally responsible for the accuracy of that advice. That responsibility cannot be delegated to an AI tool. If the advice is wrong because the AI produced a plausible incorrect answer and the lawyer relied on it without verifying, the lawyer's professional standing is at risk. The AI tool does not have professional standing. The risk transfer is entirely one-way, from the tool to the practitioner.
For professional legal practice in the UAE, this is not an abstract concern. The UAE legal profession is regulated, and professional standards require demonstrable diligence in the basis of advice. An advice memo that cannot show its sources is structurally weaker than one that can. An AI-assisted research process that produces uncited summaries creates a documentation gap that an AI-assisted process with citations does not.
Why citation-first is an engineering choice, not just a product philosophy
Building an AI research tool that cites its sources at article level requires engineering investment that is not required to build a tool that produces fluent prose summaries. You have to index the source material at a level of granularity that maps to citations. You have to associate retrieval results with their provenance and carry that provenance through the synthesis step. You have to display the citation in a form specific enough to be useful and link it to the source text so the practitioner can verify it directly.
None of these steps are technically impossible. They require decisions about architecture and investment in source indexing and provenance tracking that a tool built for speed or for breadth of coverage might not prioritise. We prioritised them because they are the product requirements for professional use. A tool that does not meet them is not suitable for the professional legal workflow we are building for, regardless of its speed or language quality.
Tariq, our CTO, spent considerable time early in development working out the citation layer architecture. The problem was not generating citations at all; a model can produce citation-looking text trivially. The problem was generating accurate citations that reflect what was actually retrieved and that point to the correct provision in the current version of the applicable statute. This required source infrastructure, amendment tracking, and a retrieval architecture that keeps the source provenance attached to the retrieved text through to the final response. It took more time to build than the response generation capability.
Why the MENA market specifically
We are based in Dubai and we built for the MENA legal market because that is the market we understand and because no one else was addressing it with the level of jurisdictional specificity that regional practitioners need. The legal AI tools that exist were built for US and UK markets. They perform poorly on UAE law questions, DIFC law questions, and Saudi regulatory questions, because the source coverage is thin and the systems were not designed around the structure of regional legal sources.
This is not a complaint about the existing tools. They are doing what they were built to do, for the markets they were built for. It is an observation that the MENA commercial legal market has a genuine gap: practitioners working with UAE, DIFC, ADGM, and Saudi sources have no well-designed research tool that covers their actual working sources. The citation requirement and the MENA jurisdictional specificity requirement are, for us, the same problem with two dimensions.
What we have built and what remains
Qanooni at early access covers UAE federal legislation, DIFC Laws and Rules, DFSA Rulebook, and ADGM Regulations. These are the source bodies that cover the bulk of commercial legal research for Dubai-based practitioners. We are adding Saudi regulatory coverage next, followed by Bahrain and Qatar. The Arabic-language source handling is something we are actively developing; our current coverage is strongest for English-language primary sources and English translations of UAE statutes. Arabic-language query handling is available, but our citation accuracy on Arabic-primary sources is lower than on English sources, and we disclose this to users.
We are building Qanooni as a team with direct experience in legal practice and legal technology in the MENA market. Anuscha spent years at the intersection of legal workflow and software before founding Qanooni. Priya practiced as a solicitor before moving into legal product work. Tariq built document-processing infrastructure for professional services environments. We understand the problem from the inside, and we are building the tool we wished existed when we were working on the practitioner side of the workflow. That is the honest origin story.
