Location: Dublin, Ireland (Hybrid) Must be located in Ireland
Role: Full-Time, Permanent
Comp: Competitive Base + Significant Equity
Visa Candidates must be Stamp 4 EU Citizen and located in Ireland no sponsorship provided
The role:
Most foundation models live in clean text or tokenized images. We build for the messier physical world: asynchronous, high-frequency, continuous multi-sensor streams running across millions of edge devices globally.
We need a Principal ML Engineer in Dublin to lead the pre-training of novel sequence and state-space foundation models from scratch on massive physical telemetry streams.
What You'll Do
- Architect and pre-train self-supervised sequence models (Time-Series Transformers, Mamba/SSMs) directly on raw, continuous multi-rate sensor dynamics.
- Run large-scale distributed pre-training jobs across multi-node GPU clusters (PyTorch, Ray, FSDP / DeepSpeed).
- Solve physical signal headaches: coordinate-frame invariance, sensor fusion, noise filtering, and temporal irregular sampling.
- Guide models from cluster pre-training through to low-latency edge/cloud execution.
What You Need
- 7+ years in ML/AI; proven track record designing and scaling deep learning architectures.
- Hands-on pre-training experience with Transformers, State-Space Models, or large sequence representations.
- Deep proficiency in PyTorch and distributed multi-GPU orchestration (Ray, FSDP, DeepSpeed).
- Experience modeling continuous physical/sensor data (IMU, accelerometers, high-rate time series, robotics kinematics).
- Advanced quantitative degree (PhD preferred, or equivalent research track record).
Visa Candidates must be Stamp 4 Eu Citizen and located in Ireland
