Member of Technical Staff, Applied Research
Sit at the intersection of model research, systems, and production outcomes for world models. Co-designing algorithms with training and inference stacks, and partnering with customers and product to train, evaluate, and deploy models under real latency, throughput and quality targets.
What you’ll do
- Advance model quality for image, video, and world-model workloads: architectures, training methods, and optimization techniques
- Co-design post-training and inference to meet production latency and cost targets
- Partner with customers and product to deliver end-to-end modeling solutions (tune → eval → deploy)
- Build evaluation frameworks for generative media and world models
- Work with performance and systems engineers to bring research into high-performance training and inference stack
- Reproduce and extend relevant literature; contribute to open-source stacks when it accelerates the product
- Identify high-impact research bets and translate them to production impacts
Minimum qualifications
- Experience building and deploying ML systems, or equivalent depth via research and open source (roughly 3+ years, or a strong applied/research track record at any level)
- Strong experience with PyTorch and modern generative architectures (Transformers and/or diffusion)
- Hands-on experience training, fine-tuning, or evaluating generative models
- Ability to turn ambiguous product or customer goals into experiments, evals, and shipped improvements
- Comfort working onsite with the team in the Bay Area
Preferred qualifications
- Depth in multimodal, video, or world models
- Experience with post-training methods (SFT, DPO, RL, reward modeling) and rigorous evaluation
- Familiarity with efficient inference and training systems enough to co-design with systems teammates
- Experience working directly with customers or design partners on applied modeling
To apply, email careers@tensorscale.io.