AristotleAI

Inspiration

Chronic muscle and joint pain costs the US $600 billion annually. Chronic pain is an interdisciplinary problem and even experts in the field don’t understand how to detect and measure pain, which has massive downstream consequences in a patient's recovery journey and pain management.

What it does

We are building an AI virtual PT assistant to 10x the value of PTs on digital health platforms. We have tech that can identify pain levels through speech and video, and we use this to personalize PT treatments, personalize encouragement during exercise regimens, and introduce holistic care.

How we built it

We used the following:

• Hume AI API to detect emotions relevant to indicate pain/relief
• Mediapipe to understand relative change in facial features
• Built deeplearning model to predict pain from facial features
• Streamlit and plotly to summarise real-time emotional response from video
• GPT-4 to summarise the emotionality of the patient through the session

Challenges we ran into

• predicting pain is an outlier event - overfitting model!
• handling states in streamlit to get the emotional responses from video
• Hume API unstable!

Accomplishments that we're proud of

Proud of the team taking on a technical challenge integrating, CV, LLMs, deep learning, real-time data vis!

What's next for AristotleAI

Website: https://aristotleai.my.canva.site/

Built With

• cv2
• langchain
• python
• tensorflow
• togetherai