The emergence of large language models (LLMs) opens unprecedented possibilities for personalised tutoring in higher education. However, creating effective AI tutors requires rigorous pedagogical configuration that most teachers do not yet master. This paper presents AI TutorForge, a web-based configuration tool enabling teachers to create personalised AI tutors through parameterisable metaprompts, grounded in a synthesis of ten major theoretical frameworks from educational science. The tool operationalises 25 tutoring dimensions across five categories: Posture and Tone, Pedagogical Strategy, Feedback Management, Question Types, and Pedagogical Context. Initial testing with teachers demonstrates the tool’s usability and its value as both a practical instrument and a reflective professional development resource. During the presentation, the author will walk participants through the theoretical principles underpinning the design, share findings from initial tests and their results, and provide a live demonstration of AI TutorForge.