Skip to content

LENS: Role-Aware Clinical Summary Grading

LENS is a role-aware multi-agent grading pipeline for clinical (ED handoff) summaries. Three clinical roles (Physician, Triage Nurse, Bedside Nurse) score summaries in parallel across 8 rubric dimensions, then an orchestrator detects disagreements, optionally adjudicates via LLM, and aggregates final scores.

Quick Start

Install the package:

pip install lens-grading

Run with the heuristic baseline (no API key needed):

lens --summary "Your clinical summary here" --engine heuristic

Run with LLM scoring (requires OPENAI_API_KEY):

lens --summary "Your clinical summary here" --engine llm --model gpt-4o

Features

  • Parallel scoring by three role-specific clinical agents
  • Shared 8-dimension LENS rubric
  • Two scoring modes: llm (OpenAI) and heuristic (keyword-based baseline)
  • Per-role weighted overall scoring based on role priors
  • Orchestrator validation, disagreement mapping, and score aggregation
  • Human-readable and JSON output formats

API Reference

See the API Reference for detailed module documentation.