The Learning Innovation Incubator at Vanderbilt

Over the summer, four LIVE students presented their research at international conferences supported by the LIVE Student Horizon Grant. This grant enables students to share their work and connect with global scholars, advancing LIVE’s visibility and impact in academic communities. Join us Wednesday to hear highlights from their presentations and experiences abroad. Pizza and drinks provided.

From Noisy Classroom Transcripts to Actionable Feedback: Fine-Tuning GPT-40 to Detect Teachers' Opportunities to Respond

Jessica Boyle 

This study examined how large language models can identify instances where teachers prompt students to participate and respond using imperfect, ASR-generated classroom transcripts. By demonstrating that meaningful insights can be drawn from noisy, authentic classroom data, this work highlights the potential for scalable feedback tools that help teachers reflect on and improve their teaching practices.

From Filling Gaps to Amplifying Strengths: Exploring an Asset-Based Approach to Learning Analytics

Fanjie Li 

This research explores how large language models can power asset-based learning analytics that help teachers notice and build upon students' contributions in K-12 science discussions. Using prompt chaining, persona prompting, and guided chain of thought, a proof-of-concept analytics prototype was developed to help science teachers notice everyday funds of knowledge students bring to the classroom and provide actionable pedagogical insights for incorporating students' home and community strengths into their teaching.

Using Whisper Embeddings for Audio-Only Latent Token Classification of Classroom Management Practices

Wesley Morris

Effective classroom management is crucial for student achievement and equity, but many teachers feel underprepared and report high stress related to behavior management. Coaching and observational feedback are effective, but they are resource-intensive and difficult to scale, especially in underserved communities. Automated feedback tools are popular, but they currently rely on text transcripts that loose information from suprasegmental features such as prosody, intonation, and affect. Our solution, MASCoT-CP, classifies teacher classroom practices directly from the audio waveforms using latent token classification, without recourse to text transcripts.

A Comparative Study on ChatGPT and Checklist as Support Tools for Unit Testing Education

Zihan Fang 

This study compares three ways to write software tests: a traditional checklist, ChatGPT, and using both together. ChatGPT made students faster and often produced better results, but many didn’t double-check what the AI wrote; the checklist helped them know when they were “done,” and many preferred it when both were available. Takeaway: teach task-specific prompting; pair checklists with ChatGPT to balance accuracy and speed; and build classroom-ready AI tools that keep students actively engaged.

November 12

AI in Legal Practice & Learning

Mark Williams & Caitlin Moon (Vanderbilt Law)

November 19

Multimodal Analytics & the Classroom of the Future

Gautam Biswas, (College of Engineering)

November 26

Holiday Break

No Session 

Explore the full lineup of Fall 2025 events at the link below