Lead Architect, Multimodal Transformers

Kodiak; Mountain View (CA); On-site; USD 230,000 - 300,000; Full time

Benefits offered by this job

Equity and annual bonuses

Flexible PTO

Office in Mountain View

Dog-friendly office

Free catered lunch

EV charging

Commuter benefits

Long Term Disability

401(k)

Job summary

Kodiak Robotics, Inc. seeks a senior technical leader to own the architecture direction of its multimodal transformer effort.

You will lead the workstream from problem framing to production deployment, and mentor engineers to raise the team’s research execution across the organization. You will define pretraining strategies for transferable multimodal representations and coordinate with perception, planning, and infrastructure teams to meet latency and compute constraints.

Qualifications

• PhD with 5+ years of industry experience or MS/BS with 8+ years.
• Lead multi-engineer technical efforts from research through production.
• Deep expertise in transformer architectures, especially multimodal or multi-stream settings.

Responsibilities

• Own the architecture roadmap for multimodal transformers fusing camera, LiDAR, and radar.
• Lead end-to-end project: problem framing, experiment design, implementation, deployment.
• Drive research direction on cross-modal attention and token fusion strategies.
• Set design review standards and mentor engineers to raise research execution.
• Define pretraining strategies including self-supervised objectives for multimodal representations.
• Partner with perception, planning, and infrastructure to align model with latency and compute budgets.

Skills

PhD 5+ yrs

MS/BS 8+ yrs

Transformer architectures

Multimodal / multi-stream

Cross-attention

Token fusion

Mentoring engineers

Python

PyTorch

Distributed training

Education

PhD in AI/CS

MS/BS in related field

Tools

Python

PyTorch

Job description

Kodiak Robotics, Inc. seeks a senior technical leader to own the architecture direction of its multimodal transformer effort.

You will lead the workstream from problem framing to production deployment, and mentor engineers to raise the team’s research execution across the organization. You will define pretraining strategies for transferable multimodal representations and coordinate with perception, planning, and infrastructure teams to meet latency and compute constraints.