T2MM: An LLM Supported Architecture For Inquiry-Based Modeling

Researchers introduce T2MM (Text to Multimodal Model), an LLM-supported architecture that generates interactive, context-aware visual models for science education rather than static images. Integrated into VERA, an inquiry-based modeling platform, T2MM outperforms traditional code-generation approaches and enables learners to adjust models dynamically, advancing how AI tools support interactive learning environments.

T2MM addresses a significant gap in AI-assisted education technology by combining large language models with interactive visualization capabilities. While LLMs have gained traction in educational settings, most implementations produce static outputs that cannot respond to user adjustments—a critical limitation for model-based learning where students need to test hypotheses and explore cause-and-effect relationships dynamically. This research demonstrates that architecture matters as much as model capability when deploying AI in educational contexts.

The development reflects broader momentum in educational technology toward interactive, responsive learning tools. As multimodal LLMs mature, institutions recognize that raw language generation lacks the contextual intelligence needed for domain-specific educational applications. VERA's ecology-based modeling domain serves as an ideal testbed because it requires both natural language understanding and real-time model manipulation—tasks that demand seamless integration between language processing and interactive UI components.

For educators and edtech developers, T2MM validates that custom architectures tailored to learning contexts outperform generic LLM applications. The procedurally generated evaluation methodology also provides a replicable framework for testing LLM-supported educational tools at scale. This approach could accelerate adoption across STEM disciplines where interactive modeling is pedagogically essential.

Looking forward, the architecture's success suggests edtech platforms will increasingly demand hybrid AI systems that blend language understanding with domain-specific interactivity rather than relying on large models alone. Institutions investing in AI education tools should prioritize context-aware, learner-responsive designs over static content generation, and developers should explore whether T2MM's architectural principles transfer to other interactive learning domains.

• →T2MM generates interactive models responsive to learner adjustments, overcoming static-output limitations of standard LLM educational tools.
• →Custom architecture designed for specific learning contexts outperforms generic full-code-generation approaches across all measured metrics.
• →The system maintains awareness of learner model context, enabling coherent, incremental modifications rather than isolated responses.
• →Procedurally generated evaluation datasets provide scalable methodology for testing LLM-supported educational applications.
• →Hybrid AI systems combining language processing with domain-specific interactivity represent the future of AI-assisted science learning tools.