العربية Start with one workflow

Studio · Khalidiyah, Jeddah · Est. 2024

Siyada Tech is a boutique AI and product studio in Jeddah, Saudi Arabia.

سيادة تك — ذكاءٌ يبقى تحت سيادتكم

سيادة

Siyada means sovereignty.

Your data stays in your tenant. Arabic is a first language, not a translation step. People make the calls that matter, and every release is measured before it reaches them.

AI that stays yours.

I Capabilities القدرات

The whole build. One senior team.

We take an AI product from first sketch to production: the interface people touch, the agents and retrieval underneath, and the evaluation that keeps it honest. Arabic and English from the first commit.

  1. 01

    Interfaces for AI products

    AI product and interface design

    Interaction models for products that think: agents, citations, review states and hand-offs, designed and engineered as one piece.

    Read more
  2. 02

    Agents that finish the work

    Agentic AI systems

    Agents that execute multi-step work inside the business systems you already run, with scoped permissions, human approval on high-impact actions, evaluation gates before release and full action logging.

    Read more
  3. 03

    Answers from your own documents, with sources

    Arabic RAG systems

    Arabic-aware chunking, hybrid lexical and semantic retrieval, reranking and inline citations over your own documents. Permission-aware, and the index never leaves your tenant.

    Read more
  4. 04

    Arabic as a first language

    Arabic NLP

    Arabic-native, not translated at the end: retrieval, extraction and generation that treat Arabic as a first language, including legal and regulatory text.

    Read more
  5. 05

    Measurement before release

    AI evaluation & assurance

    Evaluation datasets, scoring rubrics, baselines, regression tracking across releases and release gates: reporting your risk and governance committees can actually review.

    Read more
  6. 06

    Documents and processes that run end to end

    Enterprise automation

    Document and process automation across Arabic and English: intake, extraction, validation against business rules, system-of-record updates and exception routing, with an audit log at every step.

    Read more
  7. 07

    Deciding what to build, and what not to

    AI strategy & readiness

    A readiness assessment across data, systems, governance and skills; a sized portfolio of use cases; a sequenced roadmap with its prerequisites, and explicit recommendations on what not to build.

    Read more

∴ The proof in front of you

This page is part of the portfolio.

Every particle you have watched was sampled from the Siyada mark and is choreographed by your scroll, and it all runs back when you scroll up. That is the level of interface craft we bring to the products we build with you.

All capabilities

§ Charter

Built to PDPL, NCA controls and SDAIA AI-ethics principles. Customer data stays in the customer's tenant. What that means, and what it doesn't

We publish a metric only with its definition, method, sample size and date. Everything else on this site is a statement of how we work, not a scoreboard. Read the evidence policy

II Bottlenecks عنق الزجاجة

AI deployed against the step where work waits

Find the step where work waits. Release it.

AI that stays in a slide deck changes nothing. We start from one real workflow, find the step where work piles up, and ship the system that clears it: into production, then the next workflow.

  1. i — Find the knot.

    We map one real workflow with the people who run it, on real data, and locate the step where the work waits.

  2. ii — Deploy against it.

    One workflow at a time, shipped into production rather than demonstrated in slides.

  3. iii — Let agents carry the steps.

    Agents do the multi-step work inside the systems you already run, with scoped permissions, human checkpoints and a record of every action.

  4. iv — Measure every release.

    An evaluation harness ships with every system we build, so each change is measured before it reaches your people.

One workflow at a time. Shipped to production, not demonstrated in slides.

Inside your operations, AI removes the wait. Across your customer base, it becomes something you offer.

See the bottleneck patterns

III Partnership شراكة

You own the relationship. We bring the intelligence.

If you already serve customers, you already hold the hardest part of any AI business: their trust. Embed or resell Siyada's AI inside what you deliver, and capture more value from the customers you already serve.

Embed

Build our AI into the product or service your customers already rely on.

Resell

Offer our AI systems and products to your customers as part of your own portfolio.

Co-create

Build a new AI product for the market you know, with shared build and shared upside.

One partner Every customer it already serves

IV Products · proof of execution المنتجات

Our own products. Built to the standard we sell.

Built, run and evaluated by the same team that builds with you. Each one says plainly what it is not.

Plate I · Legal and regulatory research

Hukum-AI حُكم-AI

Arabic legal and regulatory research. Citation-grounded answers over Saudi law and internal policy, where every statement resolves to its source passage. Deployed in the customer's environment, with an immutable audit log.

What it is not: Legal advice. Answer quality is bounded by the supplied corpus.

Read about Hukum-AI

Plate II · Procurement and tenders

Jawabid جوابيد

Reads RFP and tender packs, extracts every requirement into a matrix, and drafts traceable responses in Arabic and English from your approved content library. Deployed in-tenant.

What it is not: It does not decide bid or no-bid, and it does not price the response.

Read about Jawabid

Plate III · Creative and commerce

Lenci لينسي

A bilingual creative and commerce AI studio: campaign and catalogue assets generated from brand guidelines and product data. Arabic is generated natively, not translated.

What it is not: Autopilot. A person approves every asset before it is published.

Read about Lenci

Plate V · Qudurat and Tahsili preparation

Jidd.ai جدّ.ai

formerly QTprep.ai

Arabic-first preparation for Saudi Arabia's Qudurat and Tahsili exams. A real diagnostic, a weakness map down to the sub-skill, a daily plan that follows the target score and exam date, and an Arabic voice tutor inside the session.

What it is not: A score guarantee. It predicts a score band; it never promises a score.

Read about Jidd.ai

?Questions

What people ask us first

What is Siyada Tech?

A boutique AI and product studio in Jeddah, Saudi Arabia. We design and build AI that works in Arabic and English, runs inside the client's environment, and is measured before release. We run our own products (Hukum-AI, Jawabid, Lenci and Jidd.ai) and build the same way for clients.

What does a boutique AI studio actually do?

We take one workflow at a time and ship a working system for it: the interface, the agents, retrieval over your own documents, Arabic language processing, and the evaluation that keeps it honest in production.

Do you work in Arabic?

Arabic and English from the first commit, including formal legal and regulatory Arabic. We measure the two languages separately.

Where does our data go?

It stays in your tenant. Corpora, indexes and audit logs remain in your environment, under your residency rules.

How do you handle Saudi regulation?

Systems are built to PDPL for personal data, NCA controls for system security and SDAIA's AI-ethics principles for governance. That is design alignment, not certification; our security page states exactly what we have and have not certified.

How do engagements start?

With a short scoping conversation about one workflow, then a focused proof on real data. If AI is the wrong tool, we say so.

Can we offer your AI to our own customers?

Yes. Companies with a customer base can embed our AI in their product, resell our products, or co-create a new one with us.

V Contact تواصل معنا

Start with one workflow.

Engagements start with a short scoping conversation, then a focused proof on a real workflow with real data. If an agent is the wrong answer, we will say so.