Non-invasive brain-computer interfaces are reaching an inflection point. Advances in neural decoding are making increasingly rich signals accessible, but the best results still rely on expensive, high-resolution systems such as MEG, while practical wearable EEG remains noisy and low-bandwidth. The opportunity is twofold: transfer what can be learned from high-quality neural recordings into cheaper EEG systems, and use modern AI to turn limited but reliable neural signals into genuinely useful communication and control.

Developing methods for using high-resolution brain recordings like MEG to inform the design and training of more practical EEG systems. These methods will be applied to build an EEG prototype combining several complementary neural signals, including intent and motor-related information. Pairing these inputs with context-aware AI that can translate limited neural bandwidth enables useful communication and control, allowing the exploration of both what non-invasive systems can reliably decode and how much real-world utility can be built on top of them.

The ARIA Opportunity Space on Scalable Neural Interfaces calls for massively scalable neurotechnologies. This project bets on extending the capabilities of existing, cheap non-invasive technologies through insights gained from more powerful, less scalable hardware and on the idea that AI can close the gap between what wearable sensors actually pick up and what's needed for useful communication and control.

Gereon studied Information Systems at TU Munich, completing his Master’s thesis at the Cyber-Human Lab in Cambridge before collaborating with the Parker Jones Neural Processing Lab at Oxford for the past two years. His work contributed to the LibriBrain Competition, built around one of the field’s most widely used non-invasive speech decoding datasets with more than 25,000 downloads. Simultaneously, Gereon helped scale enterprise AI platform meinGPT from 0 to 15,000 active users as a working student.