Language Models for Veteran Suicide Prevention (LM4VSP)

It has been called a “Veteran suicide epidemic.” According to a 2022 report, more than 22 veterans take their own lives each day which is at a rate more than 1.5 times higher than the civilian population.[1] The proposed research is to design and plan for the implementation of a trauma-informed clinical copilot powered by advanced AI to support service members and veterans with Post-Traumatic Stress Disorder (PTSD). Leveraging state-of-the-art natural language processing (NLP) models like LLaMA 2 or OpenAI's GPT-4, the clinical copilot will facilitate empathetic and secure conversational interactions. It will provide real-time emotional support and techniques grounded in best clinical practices, while adhering to best practices in trauma care and data privacy regulations such as HIPAA.The project involves the fine-tuning of AI models using domain-specific datasets, including therapeutic dialogues, trauma-informed language, and empathetic response patterns, with the input of clinical psychologists, warfighters and VA health professionals with decades of experience. Advanced safety layers, including real-time content moderation and escalation protocols, will be implemented to ensure the copilot responds appropriately and connects users to human professionals when necessary. Parameter-efficient fine-tuning techniques, such as LoRA, will be utilized to optimize performance and minimize computational costs.  [1] “The State of Veteran Suicide (2024)”, Mission Roll Call, March 26, 2024