Aadhib

Legal AI · In development

MyAdvisor.ai

An Arabic-first legal AI platform for Saudi and international business law, built on one rule — no source, no claim.

Role
Founder · Architecture · Engineering
Category
Legal AI
Status
In development
01

The problem

A general-purpose model asked a Saudi corporate law question will answer fluently and without a source. In legal work that is worse than no answer: it is an answer a professional cannot check, cannot cite, and cannot safely rely on. The failure is invisible until it matters.

02

Approach

Invert the default. The system retrieves from a curated Arabic legal corpus first, and every claim it makes has to trace to a retrieved source. Where the corpus does not support an answer, the correct behaviour is to say so rather than to generate something plausible.

03

What it does

01
Arabic-first, not Arabic-translated
The corpus, the retrieval and the interface are built around Arabic legal language rather than an English product with Arabic bolted on.
02
Coverage across business law
Legal research, contracts, litigation, compliance, company formation, governance and due diligence — the areas a business actually runs into.
03
Source-grounded answers
Retrieval happens before generation, and the source is part of the output rather than an afterthought.
04
Corpus ingestion
A pipeline for bringing legal material into the corpus in a form retrieval can actually use, which is most of the real work.
05
Billing
Built in as product infrastructure rather than deferred, because a legal platform without it is a demo.
04

The hard parts

01
Refusal is a feature
The hardest engineering problem is making a system decline confidently. A model that always answers is easy; one that knows the boundary of its corpus is not.
02
Arabic legal retrieval
Legal Arabic is precise and domain-specific. Retrieval quality depends far more on how the corpus is prepared than on which model sits in front of it.
03
Auditability
A professional needs to see why the system said what it said. That constraint shapes the architecture from the storage layer upward.

Detail

The rule the product is built around

No source, no claim.

It sounds like a slogan. In practice it is an architectural constraint that decides the retrieval design, the storage model, the prompt structure, the interface and what the system does when it does not know. Every one of those is different in a product that is allowed to guess.

Why Arabic-first changes the build

Most legal AI is built in English and translated outward. Starting from Arabic, as I did, means the corpus, the chunking, the embeddings and the terminology are all tuned for the language the law is actually written in — and the English capability sits alongside it rather than underneath it.

Status

In development. I describe no adoption, accuracy or performance here, because those are claims I would need to measure before publishing them.

Stack

Application
Next.jsNestJSExpoTurborepo
Data
PostgreSQLpgvectorRedisMinIO
Retrieval
Arabic legal corpusVector searchCorpus ingestion

Case studies

The decisions behind the build

CS-01Building legal AI that refuses unsupported claimsCS-02Architecting multi-tenant legal AI with isolation, roles and audit
OPINIONSaudi Arabia is an interesting place to build enterprise AIOPINION"No source, no claim" should be the default for high-stakes AIAI NOTEAI agents are easy to demo. Production agents are different.SYSTEM NOTEThe architecture behind a source-grounded legal AI assistant
All work