1 of 14

MindScape: Neural Odyssey

Stephen Chang, Ashmita Kumar, Russell Bustamante, and Mishty Dhekial

2 of 14

EEG

Hume

Customer

CockroachDB

Model

Training

Features

Intel

TogetherAI

Reflex

3 of 14

Our Product

Envision a reality where a simple $50 headset can transport you into a gaming experience where your emotions shape interactions with video game characters. This vision has become a reality today. We've created an EEG (Electroencephalogram) sentiment analysis pipeline that incorporates audio LLM (Natural Language Processing) techniques. It's powered by our proprietary preprocessing and modeling methods, which fuse conventional time series approaches with advanced techniques like FFT (Fast Fourier Transform) and Random Trees through multidimensional neural networks.

​

​

​

​

​

4 of 14

How does it work?

Hume API

Classify the EEG data. Each overall datapoint is labeled with an emotion according to Hume’s technology. We use this to train an EEG sentiment classifier.

EEG Data Collection

Collect EEG Data and send it through a web socket to 2 places:

  1. Reflex web app
  2. Hume API

Reflex Web App

Visualize the EEG data as it comes in real time! We draw data from the socket to help see the signals coming in from each of the 14 channels. With 120 x 14 data points coming in per second, it’s a lot of data!

5 of 14

How does it work?

Together AI

We have ascertained how the user is feeling. This is where the game comes in. The user is placed in a game where their emotion shapes interaction with the video game characters.

CockroachDB

We send the labeled data to CockroachDB. It is stored in the SQL table, and we establish a connection with Intel Developer Cloud through SSH. The data is streamed to our model.

Intel Developer Cloud

Intel Developer Cloud, with its powerful GPUs, is ideal for the training of a RandomForest classifier. We take the features and labels and fit the classifier on it. Now, when new data comes in, we predict the emotion.

​

6 of 14

Small Demo: EEG Data Visualization using Reflex

http://localhost:3000/

​

​

​

​

​

7 of 14

Hume API

Emotion classification!

​

​

​

​

8 of 14

CockroachDB

Data populated!

​

​

​

​

​

9 of 14

Intel

TensorFlow model runthrough

​

​

​

​

​

10 of 14

TogetherAI

Generates shapes for certain emotions!

​

​

​

​

​

11 of 14

The Team

Stephen Chang (Cal Poly Pomona)

  • Second Year
  • Startup Founder, Prior AWS intern
  • Experience with EEG and emotion
  • Ongoing research with Brown University

Ashmita Kumar (UC Berkeley)

  • Second Year
  • Startup Founder
  • Experience with healthcare AI/ML and databases
  • Machine Learning at Berkeley

Russell Bustamante (Carnegie Mellon University)

  • Second Year
  • Prior Amazon SCOT intern
  • CFO of Google Developer Club CMU
  • Experience in automation and software systems design

Mishty Dhekial (UC Berkeley)

  • Second Year
  • Experience in web applications and AI/ML
  • Electrical engineer on 16B course staff
  • Formula E Member

12 of 14

Next Steps

  • Create an API that can be called by third parties
  • Refine our emotion detection capabilities, introducing a wider spectrum of emotions to further personalize the gaming experience
  • Integrating AI-driven insights to provide users with actionable steps to manage anxiety
  • Community platform, allowing users to share their journeys, strategies, and progress, fostering a supportive ecosystem for mental well-being

13 of 14

Thank you!

14 of 14

Our Data:

​

[[Hume output], [[eegdataframe1], [eegdataframe2]]]