RJ·AJ
Services Veto
Riyadh · a two-person lab · governed intelligence

AI that prepares the decision. People who make it.

We are RJ and AJ, a two-person lab in Riyadh. Every system we build rests on one idea: artificial intelligence does the watching, the reading and the preparing, while a named person keeps the authority. We write the rules before the code, show the evidence behind every answer, and keep data with the people it belongs to. It is the only way we would trust a system with our own money, our own staff and our own name. The five working systems below put that idea to work in trading, strategy, HR, luxury retail and AI visibility.

▮ system readout--:--:-- UTC
gatesindependent
modelslocal first
authorityhuman · locked
5 systems · 1 trades aloneready

What we build for clients

six fields · one idea
01 · architectureGoverned AI systems

AI watches, reads and prepares; a named person keeps the authority. Rules come before code, every output carries its evidence, and data stays where it is created. This is the principle behind every system below.

02 · marketsAutomated trading systems

Trading that runs on its own, strictly within limits set before it starts: a daily trade cap, a loss stop, a Sharia screen and a first month on paper. As in ATLAS.

03 · leadershipDecision intelligence

A strategy office for leadership: the few signals worth attention, options ready before the meeting, auditable checks and a named signature. As in BOSLA.

04 · peopleAI for HR operations

An AI HR specialist that answers staff in Arabic and English, drafts documents and tracks labour-law deadlines, with a named approver for sensitive actions and employee data kept on premises. As in SANAD.

05 · commerceDigital commerce for physical businesses

A showroom taken online: buyers choose, verify their identity with the national digital ID, sign, pay, take ownership and receive delivery, with every step recorded. As in O24U.

06 · visibilityAI visibility for local businesses

Measuring whether AI assistants name a business when customers ask, with the evidence behind every answer and a clear plan to act on it. As in AIPickScore.

01 · constraint firstRules before code

Before a line is written, we decide what the system may never do, and we build that refusal as a separate mechanism the rest of the system cannot reach. Capability is easy to add. Restraint has to be designed.

02 · epistemicsEvidence, not confidence

Fluent language is not a correct answer. Every output carries its sources, its assumptions and a plain statement of what remains uncertain, so a decision-maker can disagree with it on the merits.

03 · sovereigntyPrivate by default

Data stays where it is created. Models run on the owner's machines wherever possible. We collect nothing we do not need, because the most secure record is the one that was never made.

One system, two agents, two websites

2026 · Riyadh
ATL·01
ATLAS
System · runs locally · invitation only

A trading system that works while you sleep.

ATLAS is the one system here that acts on its own. It watches the markets day and night, scores every idea and, when its gates agree, trades in your own brokerage account. Its safety lies in limits written before it ever ran: a daily trade cap, a loss level at which it stops itself, a drawdown circuit breaker, a Sharia screen based on each company's filings, and a month on paper first. Larger bets alert you on Telegram before they are placed. It runs on your machine, under your name, and stops with a single command.

runsalone · day and night
screenSharia · SEC filings
first monthpaper only
stops itselfdaily loss limit
01 / 05
react
BSL·02
BOSLA
Agent · runs locally · executive briefing

A compass for the decisions that matter.

A strategy office for leadership, not a chatbot. BOSLA distils approved sources into the few signals that deserve attention, answers strategy questions through six auditable checks (understand, gather, analyse, challenge, control and record), and prepares options before the meeting. Reversible tasks run on their own, consequential work waits for a named signature, and a kill switch outside the agents can stop everything. Missing evidence is shown as a gap, never papered over with fluent language.

checks06 auditable0
risk tiersT0 → T4
bypasses allowed000
kill switchoutside the agents
01 / 05
react
SND·03
SANAD
Agent · runs locally · in pilot

An AI HR specialist that supports the human team.

Built for a Saudi company, SANAD is a senior HR specialist that never sleeps. It answers staff in Saudi Arabic and formal English, drafts letters, tracks labour-law deadlines and supports recruitment, so a small team can serve an entire workforce. It is grounded in Saudi labour law and the CIPD Profession Map. Four categories of sensitive action require a named HR approver, it always discloses that it is an AI, and no employee record ever leaves the company's machines.

languagesAR · EN
availability24 / 70
human sign-off4 categories
employee datastays on premises
01 / 05
react
O24·04
O24U
Website · online · luxury retail, Riyadh

Luxury automobiles, personal service.

Automobile For You is a luxury and supercar retailer in Riyadh: a showroom designed around the car, the team and the operating system behind it, and a website on which a buyer anywhere in the Kingdom can choose a car with its full inspection report, verify their identity through the national digital ID, sign, pay, take ownership online and have the car delivered to the door. Money before metal, with every step recorded.

identitynational digital ID
purchasefully online
rulemoney before metal
deliveryto the door
01 / 05
react
APS·05
AIPICKSCORE
Web service · live · aipickscore.com ↗

Does AI name your business when customers ask?

AIPickScore measures whether AI assistants name a local business when customers ask. It puts the kind of questions customers in a market ask to leading AI assistants with live web search, posed as a customer in that city would pose them. It records every business each answer names and gives the owner the evidence: the captured answers, the competitors named instead, what AI gets wrong, and a plan of fixes. Saudi-first, in Arabic and English. Every number traces back to a captured answer, and being named is never presented as being recommended.

languagesAR · EN
systems measured2 · per answer
evidenceevery answer captured
rulenamed ≠ recommended
01 / 05
react
NXT·06
BUILD WITH US
Open · taking on new work · services and prices →

Local AI agents, websites, mobile apps and more.

If you need it built, we can build it, under the same rules as the systems above: rules before code, evidence behind every output, and authority that stays with you. We also run AI visibility audits and deliver AI strategy and training for leadership teams.

AI agentslocal · private
company assistantsanswer from your documents
AI customer serviceWhatsApp · web chat
AI document readinginvoices · contracts · forms
websites & storesweb apps · booking
mobile appsiOS
dashboardsmonitoring · alerts
trade huntingyour strategy · high conviction
See services and prices → Start a conversation →
01 / 05

Two people

uranium · reactor
spark → build
AJ · THE URANIUM
AJ
Sees the opening before the market names it.
energy densityhighstateunstable · in the best wayhalf-life∞

AJ is the origin of every project on this page. He walks into a room with a question nobody has asked yet and leaves with a team that cannot stop thinking about it. He reads markets, people and timing the way others read a page, and he commits early, when the evidence is thin and the upside is still intact. Energy at high density. The reaction starts with him.

opportunity hunterbrainstormuraniumenergy at high densityfirst spark
RJ · THE REACTOR
RJ
Turns the question into a system that can be trusted.
containmentnominaloutputsustainedrodsauto · adjusting

RJ is the reactor. He takes what AJ finds, contains it, and turns it into power that lasts beyond the launch: architecture, rules, code, testing and people, several workstreams at once, each finished properly. He has run a trading engine, an HR system and a luxury retail company in the same year, and the common thread is discipline: nothing leaves the reactor until it can be trusted in someone else's hands. The cooking master. The finisher.

reactorcooking masterfinishermultitasker
reaction chamberinput → containment → output
input · AJ
energy · unbounded
containment · RJ
rules · human authority
output · the company
power · sustained

One of us finds the reaction. The other contains it. That division of labour is the whole company, and it is why our systems are ambitious in what they attempt and conservative in what they are allowed to do.

How we work

method
  1. Write the limits before the code.What the system may never do is decided first and enforced by a mechanism the rest of the system cannot switch off. We treat restraint as a feature to be engineered, not a policy to be hoped for.
  2. Prove it on paper before it touches anything real.Every system earns its authority slowly, against a record anyone can inspect. Ambition in what we attempt, conservatism in what we permit.
  3. Keep the person in the loop by design.Not as a courtesy, but as architecture. The consequential step waits for a name, and the record shows whose.
  4. Speak plainly, in both languages.If a product cannot explain itself to the person using it, in their language, it is not finished. Clarity is a form of respect.

What's on your mind?

secure channel · rj·aj
channel open · awaiting input
Copied