AI Engineer (Computer Vision and Applied GenAI)

Location: Amman, Jordan. Fully remote. Type: Full-time, permanent. Level: Mid-level, roughly 3 to 5 years of hands-on experience. Languages: English required. German is a strong plus. About the role We are looking for an AI Engineer based in Amman who will work remotely with a distributed engineering team. This is a builder role, not a research role. You will take AI capabilities from prototype to something that runs reliably in production, and you will own your work end to end: the model, the data around it, the service that serves it, and the evidence that it actually works. The work spans two areas. On one side, computer vision and perception: camera pipelines, object detection and recognition, and inference that has to run fast and stay stable. On the other, applied generative AI: LLM-based features, retrieval, agents, and the evaluation harnesses that keep them honest. You will not be handed a narrow slice of either. You will be expected to move between them and to build the plumbing that connects them to a real product. Tasks Design, train, fine-tune and evaluate models for vision tasks (detection, classification, segmentation, tracking) and integrate them into production pipelines. Build LLM-powered features: retrieval-augmented generation, tool-using agents, structured extraction, and the prompt and evaluation infrastructure behind them. Write the services around the models: APIs, data pipelines, batch and streaming jobs, storage. Optimise for the target hardware, including quantisation, batching, and inference on edge devices where cloud inference is not an option. Define and track quality metrics. Establish a baseline before claiming an improvement, and be able to show where a number came from. Instrument, monitor and debug models in production: drift, latency, failure modes, and the unglamorous work of finding out why a pipeline broke at 3am. Work directly with product and business stakeholders to turn a vague need into a scoped, measurable deliverable. Document what you build so that the next engineer does not have to reverse-engineer it. Requirements 3 to 5 years building and shipping machine learning or AI systems in production. Personal projects and Kaggle notebooks alone will not cover this. Strong Python. Clean, tested, reviewable code, not notebook-only output. Practical depth in at least one of the two areas below, and working familiarity with the other: Computer vision: PyTorch or TensorFlow, OpenCV, modern detection and segmentation architectures, dataset creation and annotation workflows. Applied GenAI: LLM APIs and open-weight models, RAG, embeddings and vector stores, agent frameworks, prompt design, and systematic evaluation. Solid software engineering fundamentals: Git, code review, testing, CI, Docker, and comfort on the Linux command line. Experience deploying a model as a service and keeping it running, including cloud deployment (AWS, Azure or GCP) and basic observability. SQL and general data handling: you can find, clean and reason about the data before modelling it. Fluent written and spoken English, and the self-direction that remote work requires. You are comfortable writing things down, flagging blockers early, and working without someone checking in on you hourly. Strong plus German language skills. Part of the team and a meaningful share of the documentation, meetings and stakeholder communication are in German. Any level from solid B1 upward is a real advantage, and it will widen the scope of what you can own. It is not a hard requirement, and we will support you in improving it. Edge and embedded inference: NVIDIA Jetson, TensorRT, ONNX Runtime, OpenVINO. Video streaming and industrial camera work: RTSP, GStreamer, GenICam, machine vision cameras. MLOps tooling: MLflow, Weights and Biases, DVC, Kubernetes, model registries. Experience in an industrial, robotics, IoT or B2B product environment. A public track record: open source contributions, technical writing, or published work. Benefits Competitive salary, benchmarked to the Amman market for this level. Fully remote setup. Real ownership of features that reach customers, rather than proof-of-concept work that is quietly shelved. Direct exposure to the European market and to senior technical decision making. How we work Remote-first, with asynchronous written communication as the default and a reasonable overlap window with the European working day. Small teams, short decision paths, and direct access to the people who set priorities. We prefer a working pilot with a clear owner and a measurable outcome over a long specification. Human oversight, data protection and security are part of the definition of done, not an afterthought bolted on before launch.