I need a production-ready AI module that plugs into our existing security stack and automatically flags unauthorized taxi drivers operating at the airport. Core detection layers • Vision: reliable face recognition that picks out unique facial features, copes with the dim lighting typical of curbside cameras, and matches each face against our “authorized driver” database in real time. • ALPR: fast, accurate plate reads to link a face to the right vehicle. • Behavioural analytics: the model must learn and score patterns such as multiple daily visits, loitering around terminal doors, and repeatedly approaching passengers. Data feeds available CCTV video streams (H.264/H.265), barrier-gate vehicle entry/exit logs, and passenger incident reports are already archived and can also be streamed live; you will have API access to all three. Flow I envision 1. Video and logs feed into your model. 2. The model fuses face, plate and behavioural cues, assigns a risk score, and triggers an “unauthorised” event when the score crosses a configurable threshold. 3. Our access-control API then blocks the driver for a configurable interval (e.g., 7, 30, 90 days). Deliverables • Training and inference code (Python preferred, TensorFlow/PyTorch OK) • Pre-trained weights and a documented pipeline to retrain with new data • REST or gRPC service that returns face ID, plate, risk score and decision in <2 s/frame • Deployment guide for Ubuntu 22.04 + NVIDIA GPU • Test report showing ≥95 % precision/recall on a withheld validation set Acceptance will be based on end-to-end tests with live CCTV plus historical logs. Security and privacy best practices must be followed throughout.