AI & Machine Learning Company in Doha

Bilingual evaluation is the piece clients underestimate on Doha AI work, and it sits inside every tier below. A scoped Doha AI proof of concept runs six to ten weeks, covering data audit, a baseline model, and a hosted demo with bilingual (Arabic and English) evaluation. A custom production ML model (fraud scoring for QNB-style stacks, churn prediction, predictive maintenance for QatarEnergy rotating equipment, document extraction) adds MLOps, monitoring, and a QPDPL-aligned model card. Full production AI systems add RAG, multiple models, fine-tuning, and enterprise integrations into Cerner, SAP, Sabre, or Amadeus. Rates sit below Doha system integrator benchmarks because delivery runs from Edmonton and Chandigarh, and above offshore-only firms because we hold a Doha-based delivery lead during regulator-facing milestones. We give fixed-fee proposals rather than open T and M estimates.

Qatar passed the first comprehensive data protection law in the Gulf in 2016, and the Compliance and Data Protection Department under the Ministry of Communications and Information Technology (MoTC, now MCIT) enforces it. Our discovery phase produces a QPDPL data flow record, classifies sensitive categories (health, biometric, financial, communications, ethnicity) that require explicit consent, and maps data subject rights (access, rectification, erasure, objection). For QFC-licensed entities we apply the QFC Data Protection Regulations 2021, which align more closely with GDPR and govern data inside the Qatar Financial Centre regardless of federal law. Training data is pseudonymised before it leaves the controller environment, cross-border transfers are documented against QPDPL adequacy and contractual safeguard requirements, and inference logs are retained inside Azure Qatar Central or MEEZA. Breach notification playbooks reference MoTC and, for critical-sector clients, the Q-CERT reporting channel run by NCSA.

Yes. Bilingual Arabic and English is the default starting point, not a post-launch translation pass. We pick a foundation model based on the workload. For retrieval-heavy assistants we use OpenAI GPT-4o or Anthropic Claude through Azure or Bedrock with a bilingual evaluation harness covering Modern Standard Arabic, Khaleeji (Gulf) dialect, and English. For sovereign workloads under QPDPL or QFC residency we self-host Llama 3, Mistral, or the Arabic-tuned AceGPT and Jais families on Azure Qatar Central or MEEZA GPU instances. RAG pipelines tokenise Arabic correctly (avoiding the BPE drift that breaks naive English-only stacks on diacritised text), handle Hijri calendar references inside the retrieval index, and ship with a prompt injection defence tested in both languages. Output logs feed the client’s SIEM and an Arabic-capable human-in-the-loop review queue.

Yes. Clinical AI for HMC, Sidra Medicine, PHCC, and Aspetar runs against Cerner Millennium as the dominant EHR, with QPDPL sensitive-category handling for clinical data and Ministry of Public Health circulars governing data residency. Our pipelines keep PHI inside Azure Qatar Central or MEEZA, log every model inference for audit, redact direct identifiers before any external API call, and ship with clinician-facing explainability dashboards that surface SHAP or attention-based attributions in the EHR view. For imaging workloads we integrate against DICOM gateways and PACS deployments, with model cards that document training cohort demographics, intended use, and known failure modes for clinical governance committee review. Deployments enter through a shadow-mode phase before any decision support reaches a clinician’s screen.

Tasmu Smart Qatar is the national smart city programme led by the Ministry of Communications and Information Technology, covering transport, logistics, environment, healthcare, and sports verticals under Qatar National Vision 2030. AI projects that touch citizen services, MoCI data feeds, or government-adjacent infrastructure usually align to a Tasmu domain blueprint, which dictates open data standards, service experience patterns set by Metrash2 and Hayya, and integration paths into government-shared services. Our public sector engagements produce a Tasmu-aligned service design document, a QPDPL impact assessment, and human oversight controls proportional to the determined risk tier. We coordinate with MEEZA-hosted government landing zones where the workload requires sovereign hosting, and keep training data and inference inside Azure Qatar Central or MEEZA regions.

QatarEnergy AI work concentrates on reservoir analytics, LNG cargo optimisation, predictive maintenance on rotating equipment (compressors, turbines, pumps) across the North Field, and trading and hedging models on QSE and international LNG markets. Our OT-adjacent pipelines respect IEC 62443 zone and conduit modelling, pull telemetry through Claroty or Nozomi for ICS visibility, and never push model output back into a safety-instrumented system without an explicit human-in-the-loop gate. Reservoir and seismic models run on AKS clusters in Azure Qatar Central with GPU node pools, with PI System integration through OSIsoft connectors for plant historian data. Q-CERT ICS guidance and the NIA Cybersecurity Framework critical-sector controls are baked into the deployment, and the model card explicitly lists upstream OT segments the model is permitted to read.

A typical Doha bank model (fraud scoring on NaPS-routed transactions, AML transaction monitoring, KYC document extraction, Arabic credit decisioning) takes fourteen to twenty-two weeks from kickoff to production, assuming clean historical data and QCB Cybersecurity Framework documentation in scope. Week 1 to 4 is data audit, QPDPL classification, and baseline modelling. Week 5 to 12 is iteration, challenger model testing, and bilingual evaluation across Arabic and English customer correspondence. Week 13 to 18 is MLOps, monitoring, shadow-mode deployment, and QCB internal audit review. Week 19 onward is gradual rollout under documented kill-switch controls. Shariah board sign-off adds two to four weeks for QIB and Dukhan Bank style Islamic finance scopes. If you are pre-data or need labelled training corpora in Arabic, add six to eight weeks for vendor scoping and annotation QA.

When a project requires genuine research (novel Arabic NLP architectures, rare-event detection on LNG telemetry, reinforcement learning in trading), we scope collaborations with the Qatar Computing Research Institute (QCRI) at HBKU, the College of Science and Engineering at HBKU, the Faculty of Engineering at Qatar University, or Texas A&M Qatar rather than overselling in-house capability. Our core team handles applied engineering, MLOps, and productionisation, which is where most Doha AI projects actually stall. For standard work (RAG, fine-tuning on Llama 3 or Jais, classical ML, computer vision on known architectures) no academic partner is needed. We will tell you up front which bucket your problem fits into, and have introduced clients to QCRI’s Arabic Language Technologies group and HBKU’s Qatar Center for Artificial Intelligence when the roadmap and budget justify the engagement.