1 of 12

Better Days Ahead: A Demo

Tejas Bharadwaj, Ishant Chintapatla, Yashnil Mohanty

2 of 12

Introduction: Mental Health Issues

  • Mental illnesses affect more than one in five US. adults
  • Roughly 15 percent of teens experience/have experienced symptoms of depression

3 of 12

The Core Problem: Lack of Accessible Affordable Therapy

  • The statewide average cost for therapy is $171 per session
  • 80% of Americans surveyed, say that therapy would be beneficial to their life but 40% say they need financial aid
  • Therapy is becoming more of a luxury than a priority/necessity

​

4 of 12

Our Solution: Better Days Ahead

  • Free, 24/7, multi-modal AI Therapist
  • Powered by Chatgpt API and displayed by D-id api
    • Fine-tuned to give humane interactions
  • Uses a speech API for input and communication
  • Integrated with a face emotion detection API
    • Analyzes user’s emotions to get to know their situation better

5 of 12

Deep Dive: Chatgpt and D-id APIs

  • We utilized chat gpt 4 as our core therapist
    • The user’s input is processed by an embeddings model and a fine tuned gpt to generate a contextual appropriate and insightful output.
    • The fine tuning data as well as a chat context and a RAG knowledge base with therapy styles and transcripts allows the model to be at peak performance

6 of 12

Deep Dive: Face emotion and Speech APIs

Background Processes - Face API

7 of 12

Face-API Overview

  • Utilized the Face-API Documentation
  • Predicts facial emotions using a CV Model with percentage accuracy
  • We returned the most probable emotion variable every 5 seconds and sent the value to the Chat GPT master prompt
    • Further enhances the performance of the therapist by understanding the emotions of the user

8 of 12

Linking Face API with Chat GPT

Adds the current emotion returned by the Face-API to the Input Prompt to Chat GPT

9 of 12

Better Days Ahead Website

10 of 12

Demo Videos

11 of 12

List of APIs, Libraries and Resources Used

  • Openai GPT4
  • Openai Embeddings
  • Facerecognition
  • OpevCV to train face recognition on a custom dataset
  • Therapy transcripts libraries
  • Base-64 and requests to make a call to D-id
  • Sqllite for chatlogs and conversations across sessions
  • Gpt 3.5 for cross session summaries
  • Google native stt library
  • Elevenlabs

​

For more information about editing slides, please read our FAQs or visit Slidesgo School:

https://slidesgo.com/faqs and https://slidesgo.com/slidesgo-school

12 of 12

To Better Days Ahead