About the company
Our client does AI match analysis for youth football. Fixed cameras at the pitch record every session. The models detect the events (passes, shots, goals), track the players, and read jersey numbers off low-res footage. Academies and parents get match stats, highlights and a per-player record. This is running today at a pilot academy in Dubai, on real matches, with paying pilots. They are now taking the same stack into a second industry. First customers are committed; details in conversation. This hire owns the platform across both: making the live sports models better, and building the new product on the same foundations.
What you own
• The shared CV platform, end to end: camera ingestion, detection and tracking, OCR, classification, and the reporting pipeline. It powers a product in production today and a second one launching.
• From week one you ship to a live product: raising accuracy on the sports models with real users, while architecting the second industry's deployment on the same stack.
• Architecture: model selection and training strategy, annotation and dataset operations, evaluation methodology, the on-prem/cloud split.
• Shipping the new industry's first deployments, then hardening both products as one platform. • Technical leadership of the engineering team we build under you.
Hiring process
• Intro conversation: walk us through a CV system you built. Decisions, failures, numbers.
• Technical deep-dive with senior engineers.
• Case study on real footage: spec and prototype a detection task.
• Founders' conversation, then offer.
The whole process inside two weeks for the right person
Requirements
What we need to see
• 5+ years building computer vision systems that ran in production. Deployed systems with users, not notebooks.
• Strong Python and PyTorch, hands-on with modern detection and tracking (YOLO family, ByteTrack/DeepSORT or similar).
• Video pipeline experience: RTSP/IP cameras, scheduled or streaming inference, ffmpeg-level fluency.
• You have built datasets, run annotation operations, and measured models honestly against ground truth.
• You have fine-tuned open-source models (detection, OCR of the PaddleOCR/TrOCR class) and taken the next step: replacing them with your own models trained on validated annotations, with a working annotation-review loop.
• Fluent in the modern CV tooling ecosystem: Roboflow or equivalent for dataset and annotation operations, experiment tracking, model registries.
• Ownership temperament: you take a vague operational goal and return a working system without waiting for a spec.
Strong pluses
• OCR, action recognition or pose estimation in production.
• Sports, retail, industrial or facility camera-analytics background.
• Edge deployment (Jetson or similar) and cost engineering of inference.
• Camera and IoT literacy: sensor and lens selection, PoE networking, ONVIF/RTSP, NVRs, weatherproof housings. You can spec a site, not just a model.
Benefits
Full-time. Remote friendly: we hire the best person, not the nearest. Working hours overlapping the Gulf, with travel to Dubai for installation and milestone weeks. Competitive package plus options.