Inspiration

Our inspiration comes from a deeply personal loss. My friend's father passed away after battling Parkinson's disease, and his struggle left a lasting impact on us. For the past 11 months, we have been working on NeuroPulse AI with a shared hope: to turn our grief into something meaningful. We cannot change what happened, but we can channel our experience into research that may help us better understand the brain and inspire future neurological research. NeuroPulse AI is more than a project to us; it is a tribute to his memory and a reminder of why this work matters.

What it does

NeuroPulse AI is a research prototype that uses resting-state EEG and machine learning to investigate whether a person's baseline brain-signal patterns can predict future changes in their EEG response to repeated 14 Hz binaural-beat stimulation. It aims to explore individual differences in neural habituation, not diagnose Parkinson's disease or claim clinical benefits.

How we built it

We built the prototype using Python, MNE-Python, NumPy, SciPy, pandas, and scikit-learn. The pipeline focuses on EEG preprocessing, signal-quality checks, frequency-band power and other feature extraction, machine-learning models, and result visualization. We also explore explainable AI to understand which features influence model predictions.

Challenges we ran into

One of our biggest challenges was distinguishing a meaningful neural response from noise and individual variability. We also had to consider sampling rates, EEG artifacts, feature reliability, limited suitable longitudinal data, and the risk of misleading results when using synthetic labels. Ensuring that predictions are evaluated on unseen participants is another important challenge.

Accomplishments that we're proud of

We have stayed committed to this project for 11 months, turning a personal motivation into a structured research prototype. We are proud of developing an EEG-analysis workflow and defining a research direction centered on predicting future response changes rather than simply classifying brain signals.

What we learned

We learned that brain signals are complex, every individual is different, and machine-learning accuracy alone does not prove scientific validity. Reliable predictions require meaningful outcomes, appropriate data, careful validation, and honest reporting of limitations. Most importantly, we learned that responsible research begins with asking the right questions.

What's next for NeuroPulse AI

Our next steps are to improve feature reliability, strengthen participant-level validation, and seek suitable real longitudinal EEG data to test our hypothesis. We hope to investigate how baseline EEG relates to responses across repeated sessions and eventually compare different stimulation conditions. Our long-term goal is to contribute responsibly to neurological research while keeping our motivation rooted in the person who inspired this journey.