Overview

Application of fine-tuned large language models for healthcare decision support. Using BioBERT as the base model, this project developed an intelligent patient routing system that classifies patient cases with high accuracy — fast enough to be practical in a clinical workflow.

Key Achievements

• 91.6% accuracy on 1,000+ test assessments
• Inference optimization: Reduced latency from 16s to 5s per request
• Production-ready: Deployed on AWS Bedrock with FastAPI
• Published: IEEE paper

Why This Project

Triage decisions in clinical settings often depend on information buried in unstructured notes. The question this work tried to answer was whether a fine-tuned model could reduce that gap in a way that was fast enough to be practical — the 16s → 5s inference work came directly from that constraint.

Technologies

BioBERT · AWS Bedrock · FastAPI · PyTorch