Key findings generation with GenAI

**Executive summary:** Expel outlines a process for using large language models (with chain-of-thought and few‑shot prompting, plus RLHF) to generate standardized "key findings" from Workbench investigation data, describing input preparation, model prompting, evaluation metrics (derivative metrics, BERT/ROUGE, LLM-as-judge), and responsible-AI controls (in-house models, anonymization, PII filtering, and human-in-the-loop review) to improve analyst productivity and protect customer data.