AI Engineer

We are hiring an AI Engineer to put machine learning where it actually earns its place in our products: the assistant and routing behind Omni Channel, search and ranking across the travel platforms, document extraction in the Accounting System, and screening in HR. You take a use case from a rough idea to something running in production behind an API the rest of the platform can call — including the evaluation that says whether it is any good.

• Build and ship AI features end to end: data, model or prompt, evaluation, API, and the monitoring that follows.
• Design retrieval over our own content — product docs, tickets, invoices — instead of relying on what a model happens to remember.
• Write the evaluation before shipping, so changing a prompt or a model is measured rather than argued about.
• Serve models behind services the rest of the platform calls, inside a latency and cost budget you set and defend.
• Do the unglamorous half: data cleaning, labelling workflows, and versioning of datasets and prompts.
• Watch cost and drift in production, and act on them before someone else notices.
• Keep personal data where it belongs — anonymisation, retention, and a clear answer on what leaves our infrastructure.
• Work with the backend and product engineers so AI features are part of the product, not bolted on beside it.

• 3+ years building ML or LLM-backed systems that reached production, not only notebooks.
• Strong Python and the stack around it (PyTorch or TensorFlow, pandas, FastAPI or similar).
• Practical LLM work: prompting, function calling, retrieval, embeddings and vector stores — and knowing when a smaller classic model is the better answer.
• Comfortable with evaluation: building test sets, measuring quality and regression, and reporting the result honestly.
• Deployment experience — containers, GPU or CPU serving, queues, and keeping the bill under control.
• SQL and data modelling; you can get your own data out of Postgres without waiting for anyone.
• Able to explain a trade-off to someone non-technical without either lying or lecturing.
• Working English.

• Arabic NLP — dialect handling, RTL text, and content that mixes Arabic and English mid-sentence.
• Travel domain modelling: search ranking, pricing, or fraud.
• Speech or OCR pipelines in production.
• MLOps tooling you set up yourself (MLflow, Weights & Biases, or equivalent).
• Fine-tuning or distillation done for a measured reason, not by default.

AI here is held to the same bar as the rest of the code: it ships behind an interface, it has an evaluation, and it gets watched in production. Nothing goes live because a demo looked impressive. You sit with the engineers who own the product the feature lands in, so the trade-off between accuracy, latency and cost is a shared decision rather than a surprise.