Disentangling Prompt Element Level Risk Factors for Hallucinations and Omissions in Mental Health LLM Responses

Build a prompt-testing toolkit that validates LLM outputs against the UTCO (User, Topic, Context, Tone) framework. Developers should create an evaluation harness that specifically benchmarks model performance on high-distress mental health narratives.

Suggested repo: utco-eval

"Stop LLMs from giving unsafe mental health advice with the UTCO validation framework."

Estimated effort: 20h