INSTAR Deutschland GmbH (Waletech) Shenzhen, China
Chief Technology Officer
present
Architected a CPU-only AI inference chain: YOLO object detection → face & ANPR → fine-tuned SmolVLM scene observer. Quantised (LiteRT / OpenVINO) for edge and cloud. 10 000+ detection tasks / min in production. Built MLOps pipelines (data → training / fine-tuning → model registry → serving) with automated transfer-learning loops, version-controlled in GitLab CI/CD. Migrated a multi-GPU TensorFlow / PyTorch stack to **two CPU servers** – 10× cost reduction, equal or better accuracy. Shipped computer-vision prediction APIs and Elasticsearch-backed search for customer-facing cloud services.
