THE PROBLEM

The brain communicates through electrical signals, but reliably reading those signals outside a controlled lab setting remains largely unsolved. Existing brain-computer interface systems are narrowly trained: each new task requires a new model, new data, and weeks of retraining. EEG signals also vary significantly from

person to person, meaning a model trained on one subject rarely transfers to another. The gap between research-grade BCI and something deployable in the real world is significant: current systems are optimized for controlled conditions, not for the noise, variability, and unpredictability of everyday use. In clinical settings,

this translates directly: neurologists spend up to 30 minutes manually reviewing a single 30-minute EEG recording, hunting for epileptic spike events that last only milliseconds.

 

OUR SOLUTION

HELIN is a brain-computer interface platform built on a single pretrained EEG foundation model (LaBraM) adapted with lightweight LoRA modules. The core model stays frozen across all tasks. Each application trains only a small adapter on top, keeping compute cost low while sharing the broad EEG representations

learned from thousands of hours of diverse brain activity. This directly addresses inter-subject variability: rather than starting from scratch for each user, the foundation model already carries general knowledge of how brains work, and the adapter only learns the task-specific difference.

 

THREE MODULES

01 Motor Imagery / Rover Control

Users control a physical rover by imagining left-hand movement, right-hand movement, or a mental stop. The system classifies these three states in real time and sends drive commands over Wi-Fi or Bluetooth. No physical movement required: the interface is entirely mental, which opens the door for use cases in assistive

robotics, rehabilitation, and hands-free control in restricted environments.

 

02 Epilepsy Detection / Clinical Triage

In partnership with Rizk Hospital (IRB approved), the system processes anonymized patient EEG recordings to detect interictal epileptiform discharges, brief spike patterns that indicate seizure risk. Rather than replacing the neurologist, HELIN flags only the candidate windows. The clinician reviews a fraction of the recording instead of scanning it end to end, turning a 30-minute manual review into a targeted one.

 

03 Fear Index / Adaptive Gaming

When VR arrived, it gave games a new spatial dimension. A brain interface can give them an emotional one. This module reads arousal and fear directly from the player’s EEG and feeds a continuous index into the game engine in real time. The game adapts to how the player actually feels, not what they press. The core

challenge we tackled is separating genuine neural signal from the noise introduced by movement, facial expressions, ambient sound, and vibration. Once that filter is reliable, the interface becomes invisible, and the game becomes something genuinely new.