KASRAZEBA MULLA

Results-driven Computer Vision Engineer specializing in the design and deployment of high-throughput industrial inspection systems. Expert in architecting modular inference libraries and multi-model workflows for complex assembly verification and defect detection. Proven track record of maintaining ≥95% production accuracy and converting high-value POCs into long-term partnerships. Proficient in optimizing SOTA models (YOLO, ViT) using TensorRT and Docker for scalable edge and cloud deployments.

1. Real-Time Industrial Inspection Engine with Asynchronous Inference

• Engineered a high-throughput defect detection pipeline capable of processing 30+ FPS on high-resolution industrial streams, simulating real-world conveyor belt speeds.
• Architected a decoupled system using Redis as a message broker to separate frame acquisition from the inference engine
• Optimized model inference by implementing FP16 quantization and custom calibration, resulting in a 50% reduction in latency while maintaining >94% accuracy.
• Containerized the entire stack using Docker and orchestrated the multi-service architecture (Inference, Redis, Database) for seamless deployment.

2. Comparative Study: CNN vs. Vision Transformer for Industrial Anomaly Detection

• Conducted a comparative R&D study between CNN and Vision Transformer (EfficientViT) architectures on the MVTec AD dataset.
• Evaluated model robustness, identifying a 15% improvement in global anomaly detection using Transformer backbones.