Reducing Administrative Burden with LLM & RAG
Prior authorization requests are a major administrative bottleneck in healthcare. They require highly skilled staff to manually synthesize patient claims and progress notes into persuasive clinical letters, leading to delays in patient care and increased overhead.
We designed and built a pilot for an LLM-powered Application to automate clinical drafting:
Retrieval-Augmented Generation (RAG): The system digests raw clinical transcriptions and claims data to ground the LLM in patient-specific facts.
Citation-Based Guardrails: Every clinical assertion must cite a source document or reference established guidelines from reputable institutions. No unsourced claims allowed.
Expert Evaluation Framework: Each draft is scored by domain experts on internal consistency, clinical accuracy, and estimated chance of approval success.
Automated Triage: The system determines if a case meets confidence thresholds to proceed automatically, or flags it for expert review before submission.
Workflow Automation
Successfully automated the drafting of complex authorization letters from raw clinical notes.
Scalable Quality
Created a framework to measure and improve AI performance using expert draft comparisons.
