Representational competence in chemistry—the ability to read, draw, and translate between structural formulas and three-dimensional models—is a key prerequisite for conceptual understanding, yet it poses persistent challenges for learners. Students often struggle with conventions of different notations, translating between representation levels, and avoiding common mistakes such as octet rule violations, incorrect bonding, or misinterpretation of stereochemistry. OrChemSTAR explores how the combination of Artificial Intelligence (AI) and Augmented Reality (AR) can address these difficulties. The OrChemSTAR App, a mobile AR application that recognizes both printed and hand-drawn structural formulas, detects typical learner errors using deep-learning–based image recognition, and provides adaptive feedback by overlaying multiple representations (e.g., Lewis, wedge-dash, ball-and-stick). The app offers three modes: (i) AR mode for interactive recognition and visualization of formulas, (ii) scan mode for error diagnosis in hand-drawn structures, and (iii) practice mode with individualized exercises and feedback. We report preliminary results from a classroom study with 536 students from Switzerland, Germany, and Austria. First analyses indicate that students benefited from real-time feedback on their structural drawings and from adaptive visualizations linking 2D and 3D perspectives. Teachers emphasized the app's potential to increase practice opportunities, which are often limited by time constraints in traditional lessons. Moreover, acceptance was high across student groups, with indications that adaptive error recognition supports individualized learning paths. These findings demonstrate the potential of combining AI-based error diagnosis with AR-enhanced representations to foster representational competence in chemistry education. The OrChemSTAR App thus provides a promising approach to bridging gaps between abstract representations and student understanding in classroom practice.