AI & Machine Learning Company in Dubai

Production hardening, not model choice, is where a Dubai AI budget actually goes. All quotes itemise the 5% UAE VAT separately. A scoped AI proof of concept is a six to ten week engagement covering data audit, a baseline model, and a hosted demo on AWS me-central-1 or Azure UAE North. A custom production ML model (fraud scoring for a DIFC bank, Arabic document intelligence, demand forecasting for a Dubai retailer) adds MLOps, monitoring, and a PDPL-aligned model card. Full production AI platforms layer on RAG, multiple models, fine-tuning of Jais or Falcon, and integration with Emirates NBD or Mashreq API layers. Dubai sits at a regional premium because of the MBZUAI and G42 talent pull, but remains materially below London or Toronto for comparable scope.

The UAE Personal Data Protection Law (Federal Decree-Law 45 of 2021) and its executive regulations apply to onshore UAE entities and any controller or processor handling UAE residents' personal data, while DIFC entities follow DIFC Data Protection Law 2020 and ADGM entities follow ADGM Data Protection Regulations 2021. All three regimes require a lawful basis for processing, data subject rights handling, cross-border transfer assessment, and (under DIFC and ADGM) appointment of a Data Protection Officer for high-risk processing. For AI specifically, automated decision-making with legal or similarly significant effects triggers enhanced obligations including the right to human review. We run a PDPL or DIFC/ADGM assessment in discovery, document the lawful basis, build the data subject access workflow into the pipeline, and keep personal data inside me-central-1 or sovereign UAE cloud unless a transfer impact assessment justifies otherwise.

Yes. We have fine-tuned both Jais (Inception, G42 family, open-weight bilingual Arabic-English) and Falcon (TII Abu Dhabi, Apache 2.0 open-weight) on client-specific Arabic corpora for use cases including DIFC legal document summarisation, Islamic finance contract review (Murabaha, Ijara, Sukuk prospectuses), Emirati customer-service transcripts, and dialectal Arabic content moderation. Jais is usually the better starting point for bilingual MENA work because it was trained on a curated Arabic and English mix rather than retrofitted. Fine-tuning runs on UAE-resident GPU instances (AWS me-central-1, Azure UAE North, or sovereign G42 / e& cloud) when PDPL or sectoral guidance demand sovereignty. We benchmark against Anthropic Claude and GPT-4o on the same Arabic eval suite so you have a defensible build-versus-buy memo for your board.

Every Codazz Dubai project ships English and Arabic from day one, not as a translation pass after launch. Frontend layouts are built RTL-first with logical CSS properties (margin-inline-start over margin-left), mirrored iconography, and Arabic-Indic numeral support where the use case calls for it (Hijri calendar finance apps in particular). For NLP, we segment Khaleeji (Gulf), Egyptian, Levantine, and Modern Standard Arabic at the data layer because a single model rarely handles all four in production. Code-switched Arabizi (Arabic written in Latin script with numerals, common in WhatsApp and SMS data) gets its own normalisation step. We rely on CAMeL Tools, Farasa, and AraBERT for tokenisation and morphology, and we test against MBZUAI and TII published Arabic eval sets rather than translating English benchmarks.

Yes, and this is a meaningful Dubai specialisation. We have built Sharia screening engines that classify equities and sukuk against AAOIFI standards, contract intelligence pipelines for Murabaha and Ijara documentation, halal compliance ML for supply chain and food-and-beverage clients, and zakat calculation engines for retail banking apps. Every Islamic finance model ships with a Sharia advisory board review pack covering the underlying logic, training data provenance, and prohibited-sector exclusions (alcohol, gambling, conventional interest-bearing debt, adult content, pork). Dubai Islamic Bank, ADIB, and DIFC-licensed Islamic windows expect this artefact alongside the model card. We coordinate with your Sharia supervisory board or recommend AAOIFI-certified scholars when you do not have one on retainer.

AWS Middle East UAE (me-central-1, Dubai) launched in 2022 and now covers most enterprise workloads, with Bedrock available for Anthropic Claude and other foundation models in-region. Azure UAE North (Dubai) and UAE Central (Abu Dhabi) cover Microsoft and OpenAI workloads. Google Cloud does not yet have a UAE region, so GCP-native builds either run from Doha or are re-architected for AWS or Azure during discovery. For sovereign workloads (Dubai government, DEWA, federal entities, sensitive DHA clinical data) we deploy to G42 sovereign cloud, e& enterprise cloud, or Core42's compliant UAE infrastructure. We document residency in the PDPL record of processing activities and the cross-border transfer assessment so the UAE Data Office, DFSA, or ADGM FSRA has a clean answer if they ask.

A typical DIFC fintech model (transaction-monitoring AML scoring, KYC document extraction for Emirates ID and GCC passports, credit decisioning, fraud) takes fourteen to twenty-two weeks from kickoff to production, assuming clean historical data and DFSA reporting documentation as part of scope. Week 1 to 4 is data audit, PDPL or DIFC DP Law mapping, and baseline. Week 5 to 12 is modelling, iteration, and challenger testing benchmarked against vendor solutions. Week 13 to 18 is MLOps on me-central-1 or Azure UAE North, monitoring, shadow-mode deployment, and DFSA-aligned governance review. Week 19 onward is gradual rollout. If you are pre-data or need labelling (Arabic OCR for Emirates ID and GCC documents typically does), add six to eight weeks. We have shipped to this cadence inside DIFC-licensed stacks.

When a Dubai or Abu Dhabi project requires genuine research depth (novel Arabic NLP architectures, multimodal Arabic vision-language work, rare-event detection on Khaleeji dialect, or Sharia-aware reasoning over Islamic finance contracts) we scope collaborations with MBZUAI faculty, TII researchers behind Falcon, or G42 Inception scientists rather than overselling in-house science. Our core team handles applied engineering, MLOps, and productionisation, which is where most regional AI projects actually stall once the demo finishes. For standard work (RAG over Arabic corpora, fine-tuning Jais on client data, computer vision on known architectures, classical fraud and AML) no academic partner is required. We will tell you up front which bucket your problem fits into and which UAE research grants (Mohammed bin Rashid Innovation Fund, Dubai Future Foundation, ATRC) might co-fund the work.