Unified Vision Detection & Recognition - large scale

I’m building a production-grade computer-vision pipeline that can seamlessly switch between object detection, facial recognition, and general image classification. The model must accurately identify people, vehicles, and animals in real time, whether the camera is mounted indoors or facing unpredictable outdoor lighting and weather. I am lookng for AI specialist partners who can assist me with european based projects who are maybe in asia or americas. Here’s what I’m after: a fully trained, well-documented solution that I can deploy at scale—think edge devices for low-latency alerts as well as a cloud endpoint for heavier batch analysis. I’m comfortable with mainstream frameworks such as PyTorch or TensorFlow, but I’m open to whatever stack you feel delivers the best accuracy–throughput balance. For clarity, these are the concrete deliverables I need: • A reproducible training pipeline (data loaders, augmentation, transfer-learning strategy, and hyper-parameter notes). • Trained weights that hit state-of-the-art accuracy on the three tasks above. • An inference API (REST or gRPC) with a simple authentication layer. • Deployment scripts or Dockerfiles so I can spin everything up on both GPU servers and lightweight edge boxes. • A short read-me explaining model architecture choices, how to retrain with new classes, and expected hardware requirements. I’ll evaluate the work on precision/recall for each object class across indoor and outdoor validation clips as well as latency on an NVIDIA Jetson test rig. Once those metrics are met, we can discuss follow-on refinements like model pruning or on-device federated updates. If you’ve delivered similar end-to-end vision systems before, let’s talk and get this rolling.