Intelligent_Document_Processing_for_Graduate_Admissions__An_End_to_End_Pipeline_with_Calibrated_Abstention

Keywords: Intelligent Document Processing, Educational Technology, Human-AI, Collaboration, Calibrated Abstention, Graduate Admissions

Abstract: Graduate admissions processes face overwhelming document review burdens, with manual processing taking 15-30 minutes per application. We present an intelligent document processing (IDP) system that automates academic pre-screening while maintaining human oversight for complex cases. Our end-to-end pipeline processes scanned transcripts, resumes, and statements of purpose to extract structured academic information, assess experiential qualifications, and make calibrated admission decisions. The system achieves significant efficiency gains (70\% processing time reduction) while maintaining transparency through evidence grounding and confidence-based abstention. Experimental evaluation on synthetic data demonstrates competitive performance with GPA extraction MAE of 0.831, decision accuracy of 12.8\%, and expected calibration error of 0.691. Our modular architecture supports multiple OCR backends, configurable decision rules, and real-time processing through an interactive dashboard. This work advances intelligent document processing for high-stakes academic decision making while ensuring algorithmic fairness and human-AI collaboration.

Submission Number: 296