AI Research Architect

Location: Bay Area (Onsite or Hybrid)

Velaura AI is building the next generation of compute platforms for artificial intelligence. As AI expands beyond the cloud into edge systems and physical environments
, the constraints of energy efficiency, latency, and system architecture are becoming fundamental.

A particularly exciting frontier is physical and embodied AI
—systems that perceive, reason, and act in the real world. Applications such as robotics, autonomous mobile robots (AMRs), drones, and other intelligent machines demand new approaches to compute that combine real-time responsiveness, energy efficiency, and advanced AI capabilities.

Our mission is to design silicon and systems from first principles to enable these emerging classes of AI applications. Velaura combines deep semiconductor expertise with hardware–software co-design to deliver breakthrough performance per watt across cloud, edge, and embodied AI workloads.

We are assembling a small, world‑class team of researchers, architects, and engineers to rethink the foundations of AI compute.

We are looking for an AI Research Architect who will help shape the intersection of AI model design and next‑generation compute architectures.

In this role, you will study modern AI model architectures—including transformers and emerging alternatives—and analyze how their mathematical and algorithmic structure maps onto efficient hardware implementations.

You will help answer questions such as:

• Which model architectures are best suited for real-world AI systems such as robotics and drones?
• How should models evolve when compute efficiency becomes a primary constraint?
• How can model structures and algorithms be modified to better align with efficient silicon implementations?

This role sits at the boundary between AI research, algorithms, and hardware architecture
, and offers the opportunity to influence both the models we run and the silicon that runs them.

• Analyze modern AI model architectures—including transformers and emerging alternatives—to understand their computational structure and system requirements.
• Study the mathematical and algorithmic properties of models to identify opportunities for architectural innovation.
• Explore modifications to model structures, training approaches, or dataflows that improve efficiency on specialized hardware.
• Collaborate closely with hardware architects to translate model characteristics into efficient compute and memory architectures.
• Investigate model architectures suited for real‑time and embodied AI systems
  , including robotics and autonomous machines.
• Develop insights into how future AI models will interact with system constraints such as latency, power, and memory bandwidth.
• Track emerging research in machine learning and identify opportunities where new model architectures may benefit from specialized hardware.

• Strong background in machine learning, deep learning, or related fields.
• Deep understanding of modern model architectures, including transformers and related approaches.
• Strong mathematical foundation in machine learning, optimization, and deep learning algorithms.
• Experience working with ML frameworks such as PyTorch, JAX, or Tensor Flow
  .
• Ability to analyze model computation and translate it into system‑level implications.
• Interest in the intersection of AI models, algorithms, and compute architecture
  .

• Research experience in machine learning or AI systems.
• Experience designing or modifying model architectures.
• Familiarity with model efficiency techniques such as quantization, sparsity, pruning, or distillation
  .
• Exposure to emerging model paradigms beyond transformers.
• Interest in embodied AI, robotics, or real‑world AI systems
  .
• Experience working at the boundary between AI algorithms and hardware systems
  .

You will work on some of the most fundamental questions at the intersection of AI models and compute architecture.

Rather than optimizing existing systems, you will help define the relationship between future AI model architectures and the hardware that enables them, particularly for emerging applications in robotic…