Member of Technical Staff - Foundation Model Architecture & AI Infrastructure Vinci | Full-Time | Remote / Hybrid The Mission At Vinci, we are building the operator intelligence infrastructure that modern hardware programs rely on daily. We have already proven that a single foundation model works out of the box across industries on realistic production workloads. • Trained on 45TB+ of structured physics data • Running billion-voxel inference in production • Deployed inside Tier-1 semiconductor and hardware environments • Operating across multiple physical scales and operator regimes This is not a research prototype. This is production infrastructure. Now we are scaling deployment at industrial magnitude: • Increase simulation throughput by two orders of magnitude • Move from billion-voxel to trillion-voxel domains • Expand operator coverage across nonlinear regimes • Support global, multi-entity deployment across Tier-1 ecosystems Our ambition is not to become a frontier AI lab. Our ambition is to become the default operator intelligence layer that hardware companies run on. The Operator Frontier Today, our unified model already operates across a subset of partial differential equations in real industrial environments. The next phase is expanding that unified architecture across operators, including: • Maxwell’s equations • Elasticity • Plasticity • Navier–Stokes • Nonlinear constitutive systems • Coupled multiphysics interactions We are not building separate models per equation. We are evolving a single operator foundation model that generalizes across industries, physical scales, and conditioning regimes - and scales in deployment volume. What You Will Own This role is about AI architecture and systems engineering - not low-level GPU kernel work. You will help define and scale the core operator intelligence layer. Evolve the Foundation Architecture • Design and refine transformer variants for structured spatial domains • Explore sparse and locality-aware attention mechanisms • Build hierarchical attention across multi-resolution fields • Develop graph-transformer systems for multi-entity interactions • Improve modeling depth across nonlinear operator regimes This is architectural ownership. Scale Training & Continuous Learning • Expand distributed training beyond 45TB-scale datasets • Improve generalization across heterogeneous operator distributions • Design scalable data and curriculum strategies • Maintain reproducibility and determinism across distributed systems • Build feedback loops from deployed production environments The system must grow in capability without fragmenting in design. Architect Trillion-Scale Inference Billion-voxel inference runs today. You will help design systems that: • Scale to trillion-voxel domains • Use sparse and hierarchical computation effectively • Balance memory, compute, and communication • Maintain production-grade stability and determinism Throughput and reliability matter equally. Ship at Industrial Scale Our models already run inside Tier-1 hardware programs. You will: • Ship expanded operator capabilities into production • Increase simulations per day by 100× • Support global, multi-entity deployment • Maintain robustness under diverse industrial workloads Success is measured by adoption, throughput, and reliability — not leaderboard metrics. What We’re Looking For Deep experience in: • Large-scale foundation model architecture • Transformer variants (sparse, hierarchical, graph-based) • Distributed training systems • Production ML system design • Scaling structured datasets • Writing clean, maintainable, high-quality code You think in terms of: • Architectural generalization • Stability under nonlinear regimes • Communication vs computation tradeoffs • Deterministic distributed execution • Designing systems that become durable infrastructure You’ve built AI systems that run in production — not just experiments. Engineering Expectations • Strong software engineering fundamentals • Clean abstractions and scalable code design • Experience with modern ML stacks (e.g., PyTorch and distributed training ecosystems) • Strong CI, regression testing, and validation discipline • Comfort evolving core model infrastructure This role is about building infrastructure that lasts. Why Vinci • Single model already deployed across industries • 45TB+ structured training data • Billion-voxel inference in production • Tier-1 customers operating on real hardware workflows • High ownership at Series A stage • Opportunity to define a foundational abstraction layer early We are building something that hardware companies will depend on daily. If you want to define and scale the operator intelligence layer that industry runs on — this role was built for you.