About the RoleAs a Member of Technical Staff, Applied Research at Sieve, you'll train and evaluate multimodal models to understand how data shapes their capabilities. Your work will span video generation and audiovisual understanding, connecting advances in data curation with measurable improvements in model performance.
You'll own the research loop end-to-end: identify a model weakness, form a hypothesis about the data or training approach that could address it, build the experiment, and evaluate the results. This includes fine-tuning and post-training models, developing reproducible training and evaluation pipelines, and running controlled experiments on data quality, composition, and supervision.
You're likely a good fit if you enjoy moving between research and engineering: reading a paper, implementing a method, debugging a training run, and figuring out whether an apparent improvement holds up. You care about building reliable systems and producing findings that change how we source, curate, and use training data.
What You'll Work On
- Train and post-train models for video generation and multimodal understanding.
- Design controlled experiments to measure how data selection, mixtures, and supervision affect model capabilities.
- Build evaluations that reveal specific model weaknesses, using quantitative metrics and human judgment.
- Develop reliable training infrastructure, including distributed training, efficient data loading, checkpointing, and experiment tracking.
- Turn research findings into improvements in our data curation pipelines and products.
- Collaborate with research and engineering teams internally and at partner labs to define meaningful problems and communicate results.
Requirements- 2+ years of experience in machine learning research or engineering, with hands-on experience training or fine-tuning deep learning models.
- Strong Python and PyTorch skills, including the ability to implement, debug, and modify model training code.
- Experience designing experiments, establishing baselines, and evaluating results critically.
- Familiarity with modern generative or multimodal architectures, such as diffusion models or transformers.
- Comfortable working with large datasets and diagnosing training bottlenecks, instability, and data quality issues.
- Able to turn ambiguous research questions into concrete experiments and maintainable systems.
- Strong communication skills and the ability to explain findings, tradeoffs, and uncertainty clearly.
Bonus- Experience training video, image, or audio generation models.
- Experience with supervised fine-tuning, preference optimization, or reinforcement learning.
- Experience with distributed training and GPU performance optimization.
- Research publications, open-source contributions, or substantial independent ML projects.
- Experience as an early hire at a startup.
Benefits- 401k + Full Health Insurance
- Breakfast, Lunch, and Dinner covered and your choice of snacks
- Ubers covered home
*all roles at Sieve require you to be onsite in San Francisco 5 days per week
