Senior Multimodal AI Pre-Training Architect

Walden Robotics; San Francisco (CA); On-site; USD 280,000 - 460,000; Full time

Benefits offered by this job

Salary + annual bonus

Company equity

401(k) with company match

Flexible PTO

Daily lunch

Job summary

Walden Robotics is seeking a senior technical leader to own the pre-training of omni-models—spanning language, vision-language, video, and action—at massive scale in San Francisco. You will help set the research agenda, own architecture and data-mix bets, and push how far base models scale.

You’ll lead cross‑functional teams, balance multiple objectives, and mentor teammates while advancing training reliability and evaluation in production‑grade code.

Qualifications

• Deep, hands-on experience pre-training large models—LLM, VLM, multimodal, or VLA—at significant scale with concrete results.
• Experience training across multiple objectives or modalities and balancing them in one model.
• Hands-on experience with generative modeling of images and/or video and/or world-model approaches to physical dynamics.
• Fluency with scaling laws, distributed training, and practical failure modes of massive runs; ability to design around them.
• Track record of setting direction between research and production‑grade code, prioritizing ruthlessly under uncertainty.

Responsibilities

• Pre-Training Roadmap: Drive the pre-training agenda for omni-models across language, vision-language, video, and action.
• Omni-Model Objectives: Train a single model across losses and balance them.
• Core Modeling: Lead architecture, tokenization, and scaling-law analysis for compute allocation.
• Data & Infrastructure: Convert petabyte-scale data into training-ready mixtures with data and infra teams.
• Technical Leadership: Elevate training reliability, mentoring, and fast adoption of frontier ideas.

Skills

Large-Scale Pre-Training

Multi-Objective Training

Generative & World Models

Scaling & Systems

Technical Direction & Judgment

Job description

Walden Robotics is seeking a senior technical leader to own the pre-training of omni-models—spanning language, vision-language, video, and action—at massive scale in San Francisco. You will help set the research agenda, own architecture and data-mix bets, and push how far base models scale.

You’ll lead cross‑functional teams, balance multiple objectives, and mentor teammates while advancing training reliability and evaluation in production‑grade code.