SCRIBE: A Fine-Tuned Transformer Embedding Model for Evaluating Medical School Personal Statements

Steven Lawrence Fernandes

Keywords: Transformers, Embedding, Fine-tuning

Abstract: Personal statements are a crucial part of the medical school application process, offering applicants the opportunity to showcase their personality, experiences, and motivation for pursuing a career in medicine. However, many students struggle to draft and revise these essays while juggling pre-med commitments. To address this, we present the Semantic and Contextual Rubric-Based Intelligence for Biomedical Essays (SCRIBE), a novel offline tool for the automated evaluation of medical school personal statements, which fine-tunes the state-of-the-art e5-large-v2. SCRIBE provides automatic, structured feedback, lowering barriers for students who may lack access to mentoring or editing support. Our tool segments essays into semantically coherent sections, classifies each into rubric categories (Spark, Healthcare Experience, Showing Doctor Qualities, Spin), and assigns a score from 1 to 4. The SCRIBE tool was trained on manually annotated text by subject experts. The novel contributions of our research work include: (1) Development of an automated tool named SCRIBE for practical evaluation of medical personal statements by fine-tuning the state-of-the-art e5-large-v2 model, which is publicly available.

Submission Number: 367