Decoding Brain Activity using Machine Learning
San Francisco, California, USA
2011
Bio-Informatics Startup
~100
Machine Learning Intern
At Emotiv, I worked on developing foundational encoding models for EEG data, specifically focusing on enabling joint learning across heterogeneous datasets—differing in sampling rates, gain levels, and electrode configurations.
After surveying and replicating leading state-of-the-art EEG transformer architectures, I collaborated with the research team to propose novel candidate models that incorporated spatial and frequency embeddings. These models were tested using both supervised and contrastive (unsupervised) pretraining, allowing us to learn generalizable representations from Emotiv’s diverse internal data.
Once pretrained, our models were evaluated across internal clinical and consumer-facing EEG classification tasks—showing significant improvements in performance over existing baselines.
In parallel, I contributed to Emotiv’s clinical trial infrastructure by building an automated AWS pipeline. Using S3, Lambda, and MFA-secured functions, I designed a system to generate audio from text passages and perform fine-grained, word-level alignment between the spoken and written forms. This was key for synchronizing auditory stimuli with EEG recordings in tightly controlled experiments.
This role pushed me to integrate deep learning research with real-world deployment constraints, while contributing directly to the company’s mission of making brain-computer interfaces more scalable and clinically viable.