About the job
Hello, I'm Brian, the Co-Founder of Egra. We have recently secured $5.5M in funding to develop foundational models for brain signals, and we are on the hunt for our inaugural research engineer.
From day one, you will possess full ownership of your projects. Forget about lengthy onboarding processes, waiting for approvals, or navigating through bureaucracies. Here, you will collaborate directly with the founders on complex technical challenges, armed with the resources necessary to address them. You will define the architecture of our infrastructure, make pivotal engineering choices, and construct the systems that drive our research forward. If you excel in an environment that values high agency and wish for your contributions to directly influence the company's direction, this is an exceptional opportunity.
Key Responsibilities
EEG, or electrical activity of the brain recorded from the scalp, is one of the most challenging real-world signal modalities in machine learning due to its low signal-to-noise ratio, significant subject variability, and inconsistencies across devices. Most shy away from these challenges.
As our founding research engineer, you will oversee the systems that facilitate our research. Specific projects you will manage include:
Creating versioned and reproducible preprocessing pipelines for EEG data from various sources, including device-specific normalization, channel mapping across montages, artifact detection, and signal quality assessments. Your systems will be able to instantly respond to queries like, "Which preprocessing version generated this result?"
Designing infrastructure for experiment tracking and training that allows us to conduct numerous pretraining experiments concurrently without losing track of changes in hyperparameters, data splits, preprocessing versions, and model checkpoints, ensuring everything is linked and reproducible.
Developing a data ingestion system capable of assimilating various EEG formats (EDF, BDF, BIDS, proprietary device exports) and normalizing them into a clean, internal representation.
Enhancing training pipelines for optimal throughput on noisy, variable-length signal data using mixed precision, intelligent batching across different recording lengths, and efficient data loading for datasets that don't conform to standard loaders.
Our Vision
We are aspiring to create a world where thought serves as an interface.
Imagine composing a message silently as it types itself. Or navigating an AR display without any physical interaction. Our software aims to adapt to your cognitive state in real time, establishing a universal interface between human thought and digital interaction.

