Intern

Egra•New York, NY

4d•$125,000 - $300,000•Onsite

About The Position

This posting is for a single role with wide possibilities, accommodating full-time researchers, full-time engineers, and interns. The company emphasizes complete ownership from day one, with no lengthy onboarding, waiting for permission, or navigating layers of approval. The work involves tackling real problems with real compute and real autonomy, with contributions expected to be integrated into the product within the first month. The core focus is on training models using high-bandwidth, multimodal data including biosignals, eye-tracking, pupillometry, video of human faces, audio, content humans are reacting to, and downstream behavior. The role blurs the lines between research and engineering, requiring individuals who can both conceptualize and implement solutions rapidly.

Requirements

Ability to take a vague research direction and ship something concrete within a week
Strong, experiment-informed opinions about what makes representations generalize
Comfort with heterogeneous, multimodal data and a toolkit for making it useful
Proficiency in shipping fast with AI coding tools (e.g., Codex, Claude Code, agents)
Demonstrated 'taste' in evaluating benchmarks and model architectures
General modeling ability on hard data; domain expertise in specific signal modalities is not required
Ability to work autonomously without a clear roadmap or constant managerial guidance
Interest in building systems that work in production over theoretical neuroscience

Nice To Haves

Experience with EEG or a neuroscience background (Note: This is explicitly stated as NOT a requirement, but could be considered a 'nice to have' if the candidate also possesses strong ML skills)
Experience with hand-crafted features, classical signal-processing pipelines, or domain-specific engineering as a primary contribution (Note: This is explicitly stated as something NOT to apply if it's your main contribution, so it's not a 'nice to have' in the traditional sense, but indicates areas of potential interest if balanced with ML skills)

Responsibilities

Designing self-supervised pretraining objectives on multimodal physiological + content data
Stress-testing recent multimodal / signal foundation model papers to understand limitations under distribution shift
Building evaluation protocols that distinguish real progress from benchmark noise
Shipping internal research tooling such as experiment tracking, dataset versioning, and agentic eval pipelines
Closing the loop between offline model results and the live product
Writing internal research memos that form the shared knowledge base, documenting findings on model performance and experimental outcomes