Software · Infrastructure · Arabic AI

Systems people work in.Intelligence inside them.Servers that carry them.

Kindi designs, builds and runs all three, in Saudi Arabia.

  • Arabic first, in both languages
  • A person approves every AI step
  • Hosted on our servers in the Kingdom
  • Staff portals
  • IT support desks
  • Electronic forms
  • Contract registers
  • KPIs on the org chart
  • Beneficiary services
  • Photo archives
  • Oral-history archives
  • Single sign-on
  • Nafath sign-in
  • Access profiles
  • Web platforms
  • Local payment gateways
  • Dedicated servers
  • GPU model hosting
  • HPC and job queues
  • Printed Arabic OCR
  • Handwriting recognition
  • Speech to text
  • Arabic search

How a project runs

From a login to the servers.

We build in the order people meet a system: what they log into first, then everything that carries it.

  1. 01

    Scope

    What people will log into, and who is allowed to see what. The scope is agreed before any work starts.

  2. 02

    Portal and apps

    The service takes shape: one portal, and the apps behind it, Arabic first and in both languages.

  3. 03

    One backend

    One backend with identity and single sign-on, so a single login opens everything a person may use.

  4. 04

    Servers

    It runs on machines we operate in Saudi Arabia, and we keep it running there.

Kindi Labs

Kindi Labs: models that read Arabic.

We build models that read printed and handwritten Arabic, test them on fixed exams, and publish the numbers.

  • 0.89%Kindi OCRPrinted Arabic: character error on our test of 90 blocks, 4,320 words
  • 8.83%Kindi HTRHandwriting: character error on our test of 2,000 lines, 24,453 words
  • 87.1%laya-ara-ragArabic retrieval, first pick right on SadeemQuestion
Visit Kindi Labs
A manuscript page from our handwriting test: British Library, Or 13006. Each line is found first, then read.Public domain, British Library

Infrastructure

On our own servers, in Saudi Arabia.

The systems we build, and the models inside them, run on hardware we operate ourselves. We don't resell someone else's cloud.

  • Dedicated serversFor client systems and their databases.
  • GPUs for the modelsReading, speech, and search models run on our own GPUs.
  • Job schedulingHPC management and job queues.
  • Data stays in the KingdomData stays on our servers inside Saudi Arabia.
Infrastructure
An illustration of the scheduler placing work on GPUs.