Authors:

Ha Nguyen

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Modern educational systems struggle to offer personalized, inquiry- based learning that develops critical thinking. Traditional teaching often falls short in engaging students through guided questioning, limiting their ability to understand concepts independently. Socratic AI tutors combine the proven pedagogical effectiveness of Socratic questioning with the scalability of artificial intelligence, delivering structured, question-driven support instead of direct answers. However, in specialized fields like Biological Engineering, existing tutors lack in-depth domain expertise and depend on costly, closed-source models. To address this gap, an open-source Gemma-7B language model was fine-tuned and integrated with a retrieval-augmented generation (RAG) pipeline and Socratic dialogue style, creating a custom Biological Engineering AI Tutor. A hybrid synthetic dataset of Socratic teaching dialogues was used for fine- tuning, and a knowledge base curated from over 15 textbooks, peer- reviewed articles, and course materials powered the retrieval component. Performance evaluation, measured by accuracy, relevance, coherence, and inference latency, demonstrated that the custom tutor matches commercial LLMs in accuracy and content relevance, maintains interactive-speed responses, and operates at a fraction of the cost. User feedback indicated strong student engagement, with learners affirming that Socratic questioning promotes active learning. This research established a scalable, cost-effective framework for specialized engineering education. Future steps include structured evaluation of student outcomes, direct comparisons to commercial models, integration into classroom settings, iterative refinements to validate learning gains, and adaptation for other disciplines. Keywords: LLMs; Socratic Teaching; Biological Engineering; Synthetic Data Generation; Retrieval-Augmented Generation † Presenting Undergrad Author; ‡ Contributing Undergrad Author; * Undergrad Acknowledgment

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Ha Nguyen