Senior Machine Learning (ML) Engineer

Startup. We're building superintelligence to reclaim your attention and focus. Attention is how you navigate the world, learn, form memories; it’s how you experience reality. Today, our attention has been hijacked by algorithms designed to extract data from you to feed Big Tech business models.

Our team is building tools that let you see, protect, and own the most important resource you have, attention.

We’re starting with a Mac app that collects behavioral signals and usage patterns across a person's devices - digital phenotyping for cognitive states. We use on-device ML to classify mental states and activities. This enables new types of insights about how you think and work (Oura Ring for your mind) while laying the foundation for a new generation of attention-preserving adaptive interfaces powered by local models.

Real ML, not just LLM wrappers. We're training temporal models and random forests on behavioral data.

Access to real data. We're collecting high-fidelity behavioral signals with a clear path to training novel models.

Lots of compute. You will have access to two NVIDIA Sparks and practically unlimited credits for API calls to frontier models.

Privacy-first architecture. Everything runs on-device. We're exploring federated learning for model improvement without centralized data.

Small team, high ownership. You'll work in a pod of 4. No bureaucracy, no waiting for approvals. Ship, communicate, iterate.

Mission that matters. We're building tools to help people reclaim their attention from systems designed to exploit it

Right now, parts of the ML workflow are still too manual (e.g., data export → model runs → results). We're hiring a senior engineer to join our team to make the Python-side ML system reliable and easy to iterate on: data processing, feature pipelines, on-device inference, training pathways, evaluation, and privacy constraints. For local training and inference, you will have access to two NVIDIA DGX Sparks.

• Interview process that respects people and their time
• Professional and open IT community
• Internal meet-ups and resources for knowledge sharing
• Time for recovery and relaxation
• Bright online and offline events
• Opportunity to become part of our internal volunteer community

• Build the Python-side ML pipeline: ingestion → normalization → feature extraction → inference/training → evaluation artifacts. Initially backend-focused, increasingly ML platform work.
• Make privacy real in the ML workflow. We're starting with Flower for federated learning.
• Drive model selection for behavioral classification, prediction, and other ML tasks based on feature needs. Through research, experience, and/or prototyping and benchmarking, determine what actually works for our data and constraints, and make the
  call on what ships.
• Design the training pathway so user-labeled feedback and behavioral signals become training-ready datasets (repeatable, documented, testable).
• Implement real-time classification where ML models receive behavioral data from the Swift app and return predictions in-app.
• Improve iteration speed: better evaluation loops, dataset tooling, and small automation wins that let the team move faster.

Must-haves

• Appetite and capability to flex between Python backend and ML platform and ML work Comfort training and iterating on models — you don't need to be a pure researcher, but you understand what training and evaluation require 
• AI-forward workflow: you use Cursor, Claude, or similar tools to move faster
• Strong systems integration instincts (contracts, schemas, versioning, failure modes) even if you're not writing the client
• Startup experience or mindset: comfort with ambiguity, high agency, fast iteration

Nice to have

• On-device ML experience (runtime constraints, packaging, performance profiling)
• Background or interest in cognitive science and attention
• Experience building data/ML pipelines
• Privacy-preserving ML patterns and/or federated learning familiarity
• Time-series or behavioral modeling experience
• Experience integrating ML into product — you've pushed ML across the boundary into something user-facing (even if your prior "UI" was an internal tool)