Staff AI Architect: Scaling Large Multimodal Models

Hark; San Jose (CA); Hybrid; USD 180,000 - 450,000; Full time

Job summary

Hark in San Jose is seeking a Member of Technical Staff in Architecture & Scaling to shape the training of our largest models. You will design and refine training regimes, select architectures, and ensure scalable, reliable performance as we push to new compute frontiers.

This role blends empirical research with production impact, requiring hands-on implementation, rigorous experimentation, and collaboration with multi-disciplinary teams to advance multimodal AI hardware and software.

Qualifications

• Hands-on experience training large models at scale.
• Strong empirical instincts for experimental design.
• Fluency with scaling laws and using small-scale results for frontier-scale bets.
• Deep familiarity with modern training stacks and distributed training on large GPU clusters.
• Strong engineering; research requires coding and profiling.
• A track record of shipped work in papers, systems, or production.

Responsibilities

• Research model architecture, optimization, and scaling to improve the capability and efficiency of large models.
• Establish baselines and run controlled experiments to test scalability for frontier training.
• Set training recipes at scale: learning-rate schedules, context length, token budgets, compute allocation.
• Explore new architectures including mixture-of-experts and long-context extension.
• Diagnose instability in large runs: loss spikes, divergence, numerical issues, and infra failures.
• Work across modalities since models are multimodal from pretraining forward.

Skills

Training large models

Empirical experimentation

Scaling laws

Distributed training

GPU clusters

Strong engineering

Job description

Hark in San Jose is seeking a Member of Technical Staff in Architecture & Scaling to shape the training of our largest models. You will design and refine training regimes, select architectures, and ensure scalable, reliable performance as we push to new compute frontiers.

This role blends empirical research with production impact, requiring hands-on implementation, rigorous experimentation, and collaboration with multi-disciplinary teams to advance multimodal AI hardware and software.