AI automation that shows its working

Agents that do named jobs — triage, chase, draft, score — and print the reasoning, because "the AI decided" is not an answer you can act on.

The useful question is never "can we add AI". It is which decision, made how many times a week, by someone reading the same four fields every time. Those are worth automating, and they are worth automating visibly: every agent we ship prints why it chose what it chose, so a supervisor can overrule it and a manager can audit it.

What you get

  • Triage that reads an inbound message and picks the trade and the priority
  • Scoring with the weights visible, so a human can disagree on the evidence
  • Escalation ladders that fire on a clock, not on someone remembering
  • Drafting in English and Arabic, always reviewed before it sends

In more detail

What the agent is not allowed to decide

Every agent runs inside a boundary drawn before it does. In JD Property OS the repair-or-replace call weighs its signals, but an asset still under warranty overrides the weighing outright — the rule wins, not the score. In the ComplyGCC MVP the roles are enforced in every server action, so an auditor cannot write no matter what an agent suggests. That boundary is named in the written scope, next to what deliberately stays manual.

What the model is allowed to see

Less than people assume. The ComplyGCC MVP strips every piece of personal data before anything reaches a model, and its report drafts come back with placeholder tokens that are substituted on your own machine. Because it keeps a single data layer, moving where the data sits is a one-line change rather than a rebuild. The same build names which parts are simulated — the watchlist, the FATF snapshot, the filing.

Why the timing lives in a tested module

An agent that acts on a schedule is only trustworthy if the schedule itself can be tested, which means the timing rules cannot live inside a screen. JD Property OS keeps its decisions in 7 pure modules, with 63 assertions written against the operation. Frozen means what it says: the deadline is fixed to the moment the message landed, not recalculated as the job moves.

The input is not always typed, and not always in two languages

Warehouse staff do not always share a language with the finance team. The Atlas ERP sales demo takes a WhatsApp voice note in Urdu and parses it into a goods-received note that really posts — and since every action there writes a balanced journal, it arrives as accounting rather than as a message somebody still has to key in. Atlas is a sales demo for GCC construction, not a system running anyone's company.

Proof, not a promise

Common questions

What does "shows its working" look like on a screen?

A number with its inputs printed beside it. JD Property OS scores technicians on four named values — trade, distance, workload and shift — so a supervisor who disagrees is arguing with those four values, not with the software. Where the decision is a rule rather than a score, the rule is named too.

Does our customer data get sent to an AI model?

Only what the job needs, and often nothing personal at all. ComplyGCC is built so that zero personal data reaches the model: it is stripped first, and the placeholders come back to be substituted on your machine. Which fields a model may read is written into the scope before anything is built.

What happens when the agent gets it wrong?

It costs you a correction. The agent fills one named field — the trade, the priority, the score, the draft — and prints the reasoning beside it. The blast radius is one field with its working shown, not a decision taken silently somewhere you cannot see.

How is AI automation quoted?

Scope decides it, so we quote per project — which decisions get automated, and what they have to read, change the work. You get one written figure after a call, and it changes only if the scope does.

Tell us what is breaking How we quote

It boots in your browser and runs the real logic — no form, no sales call.

What we build

JD Software Solutions — a four-person software studio in the United Arab Emirates. We build the system, hand over the code, and stay on to run it.