2026 ASEE Annual Conference & Exposition

The increasing availability of generative artificial intelligence has raised concerns about the authenticity of typed student writing, particularly for Writing-to-Learn (WTL) activities intended to capture students’ learning processes rather than polished final answers. Handwritten responses provide a potential means of preserving authenticity, but their instructional use at scale is limited by the difficulty of digitization and analysis. This paper investigates the feasibility of using an AI-based handwriting-to-text transcription pipeline to support handwritten WTL artifacts in undergraduate engineering education. The study is situated within a multi-cycle WTL pedagogy and focuses on three reflective learning processes: learning plans, learning evaluations, and learning reflections. Handwritten responses were collected from junior- and senior-level engineering courses and digitized using a cloud-based vision–language transcription pipeline with human-in-the-loop validation. Quantitative transcription accuracy was computed at the word level by comparing system-generated transcripts with gold-standard human reference transcripts across sampled student submissions, while submission-level structural checks were used to examine attribution, page grouping, section association, and instructional usability. Results indicate that the proposed pipeline consistently produces accurate and semantically faithful transcripts of handwritten reflective writing, outperforming baseline optical character recognition (OCR) approaches and a naive vision–language model baseline under authentic classroom conditions. These findings suggest that AI-assisted handwriting transcription can enable scalable use of handwritten Writing-to-Learn activities while preserving pedagogical intent and authenticity. The work contributes evidence that handwriting, supported by appropriate AI tools, remains a viable response modality for reflective learning in the generative-AI era. In addition, by evaluating the system on authentic HBCU classroom data, this study helps address the underrepresentation of such educational contexts in handwriting-recognition research and provides initial insight into the robustness of AI-based transcription in diverse educational settings.