Serra Vision

Serra Factory's Serra Vision project.designed to detect objects like forklifts, persons, and drivers across their steel manufacturing environment.had been brought to a halt by four compounding technical challenges. First, system instability: the existing system was plagued with bugs and performance issues that caused frequent crashes, making object detection unreliable and the platform unfit for production use. Second, outdated AI technology: the AI models powering detection were obsolete, resulting in lower accuracy and significantly slower processing times that couldn't meet the demands of a live industrial environment. Third, integration difficulties: attempts to introduce new features and technologies into the existing system created ongoing delays and technical obstacles, making it hard to move the project forward without breaking what was already in place. Fourth, a lack of advanced capabilities: the system had no real-time incident recording, no detailed visualization tooling, and no configurational logic.limiting both its practical effectiveness and its commercial appeal to potential new clients.

CodeBricks embedded directly with Oscar Serra's team to first stabilize what existed before building on top of it. A thorough bug audit was conducted to identify and resolve all critical issues causing system crashes and unreliable detection. Once stable, the AI pipeline was redesigned: YOLO models were introduced for high-accuracy object detection, DeepStream was integrated for superior real-time data processing across video streams, and CUDA acceleration was applied to dramatically improve inference speed. With a reliable, high-performance foundation in place, the team layered in the advanced features Serra Vision needed to be a competitive commercial product.Virtual Twin visualization, configurational detection logic, and a full incident recording and playback system integrated into a web-accessible interface.