This role is with one of Dex’s trusted partner companies. We work closely with their teams to truly understand their culture, goals, and what they’re looking for, so we can match you with the right opportunity and give you context about the role before you commit to a process.
The role
This early-stage, venture-backed company is building foundation models for extreme physics. Think semiconductors, aerospace, defence, and fusion energy. Existing simulation tools are too slow, too brittle, or too expensive to keep pace. Their mission: accelerate progress in fields critical to global infrastructure and energy.
You’ll be the first dedicated engineering owner for the ML stack. This isn’t a narrow systems role, and it isn’t pure research. You’ll join a small, research-heavy team, owning everything from model code and training workflows to experiment infrastructure and repo standards. You’ll also build distributed training capabilities and the backend platform that delivers models to customers. This is a rare seat: you’ll set the engineering culture from day one, with genuine latitude to contribute to model development itself.
The work
- Turn research prototypes into robust, production-ready ML code. Integrate disparate research branches into a single, coherent codebase.
- Profile and resolve bottlenecks in training, ensuring stability and speed for large-scale models.
- Architect and implement distributed training capabilities from scratch, enabling multi-GPU and multi-node setups.
- Build critical in-house tooling for hyperparameter optimization, ablation, and experiment tracking.
- Own the backend platform that delivers these foundation models to customers.
What You Bring
- You’re a senior, hands-on engineer who still writes and ships critical code, thriving with high agency in an early-stage environment.
- Deep PyTorch experience across model code, data pipelines, training loops, and runtime behavior – not just at the modeling layer.
- Practical experience with distributed training (multi-GPU/node, GPU clusters), including memory, communication, and stability challenges.
- Proven model-engineering background in physics/simulation, vision, or LLM systems, with a track record of improving ML systems through measurement and profiling.
- Strong Python platform and backend foundations: API design, workflow orchestration, and operational tooling (Docker/K8s/IaC).
#J-18808-Ljbffr…