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.