Developed computer-vision algorithms in Python (PyTorch, TensorFlow, OpenCV), using CNNs and Vision Transformers (ViT).
Solved the "car leaving a parking spot" challenge by combining deep-learning detection with classical CV to cut false alarms — reaching 84% precision with only a 6% recall trade-off.
Built high-performance RESTful APIs in Python (FastAPI) for real-time data processing and seamless client-server communication.
Integrated computer-vision models into the backend pipeline, connecting the AI detection models with the web application.
Developed the system frontend using Next.js, React, and TypeScript.
Built a custom image-tagging and model-training platform, plus AI agent tools and infrastructure, to speed up training and solve complex edge cases.
Accelerated development and coding workflows using Cursor, Claude Code, and MCP.
Almog Arazi
