Psiops

Inspiration

We know from experience how frustrating, slow, expensive, and difficult receiving psychological care can be.

This project is a proof of concept for AI assistants to perform intake and administer preliminary psychological care for people in crisis situations -- where time is of the essence, yet human resources are often limited. The goal of this is not to suggest replacing human involvement in mental health-care, but rather to equip existing care providers with a tool to intake, triage, and help intervene more efficiently.

What it does

Receiving psychological care is a slow, expensive, and frustrating experience. This is less than ideal -- especially for such a critical service. This project seeks to leverage LLM's natural language understanding capabilities to administer preliminary psychological care, and analyze and report on the user's psychological state, much like how a first responder administers first-aid and informs the emergency staff on the status of the patient.

There are two components to this project -- the chat screen for users, and a dashboard for care providers.

The workflow is as follows:

1. User chats with an AI assistant about their mental health. The assistant will aim to steer the conversation towards improving or stabilizing the user's state. The user is always provided with an option to speak with a human counselor at any time with a call button.
2. In the admin dashboard, the conversation with the user is summarized and analyzed for risk of harming oneself or others. Emotional states of the user throughout the conversation is tracked and graphed as a spider chart.
3. Intervention level is logged and displayed. The AI assistant will automatically escalate intervention levels every time it detects a need for human intervention (increases in risk of harming oneself or others) and alert the supervising human through the dashboard.
4. A mental health professional reviews the user's chart and determines an intervention approach.

How we built it

We combined our team's expertise in ML and software development. Furthermore, we leveraged GPT's NLU capabilities to generate specific tokens under certain circumstances that would automatically escalate risk levels (e.g. the user shows signs of increasing destabilization)..

What's next for Psiops

We would like to extend our system's capabilities to support automatic transcribing and note taking during virtual or in-person sessions with a mental health professional. We would eventually like to create a fully managed Practice Management System for mental health professionals.

Built With

• docker
• javascript
• node.js
• python
• react
• redis
• typescript