Research Engineer, Post-Training

Vizcom • San Francisco, CA

3d • $250,000 - $450,000 • Onsite

About The Position

Vizcom is a Series B company that has raised over $52M from investors like Radical Ventures and Index Ventures. Our platform is used by design teams at major companies such as Nike, GM, New Balance, and Hasbro for daily design tasks including sketching, rendering, color and material selection, 3D modeling, and exporting. We focus on the 'pencil to product' process, aiming to understand the critical design decisions made by professionals that lead to a final product. We possess a unique dataset of millions of design moments, and our first research challenge is to understand the judgment embedded within these decisions, which current reward models cannot replicate.

Requirements

• Strong programming skills and ability to write trustworthy ML code.
• Experience with post-training diffusion or flow models, including supervised fine-tuning, preference optimization, or RL.
• Excitement about a product-coupled research approach with audit-grade rigor.

Nice To Haves

• High-performance implementations of training or inference code.
• Experience making physical things, or a deep care for those who do.
• A disposition focused on how professionals decide, rather than what the internet likes.

Responsibilities

• Build models that can understand the judgment that carries a design from pencil to product.
• Work alongside the engineer who built the current post-training stack and the research log.
• Own the post-training roadmap for models, from research plan through production.
• Conduct reward and preference modeling on years of professional design decisions.
• Implement the full method stack including supervised transfer, distillation, and reinforcement learning.
• Define and maintain research standards, including the research log, evaluations, and the criteria for earning compute.
• Manage the feedback loop with product to shape future tool development based on current signal limitations.
• Evaluate new developments in post-training and integrate promising techniques into the existing stack.

Benefits

• Access to a unique dataset of five years of professional design decisions, continuously growing.
• Significant compute resources.
• Models deployed to users within weeks, not quarters.
• Direct line of communication to founders, including the CEO who has a background as a transportation designer.
• Visa sponsorship available.

What This Job Offers

• Job Type: Full-time
• Career Level: Senior
• Education Level: No Education Listed