Research Engineer - Multimodal

About the role

We’re looking for scrappy and self-motivated people who have full-stack machine learning skills: collecting data, training state-of-the-art models, building evaluations, writing efficient inference algorithms, and iterating on user feedback. In the day-to-day, you will be responsible for developing new AI capabilities end-to-end. This means you will need to wear a lot of hats across the full ML stack. You should be comfortable thinking about all parts of the problem, and ready to work on any and all components of it.

• Determining the type of training data we need, finding where we can collect it, and writing distributed data gathering pipelines to ingest data
• Developing new model architectures that push the state-of-the-art in terms of quality, scale, and inference speed
• Creating new evaluations that capture different aspects of generative outputs
• Writing fast inference algorithms to serve these models at scale
• Working with product teams to integrate feedback mechanisms into the product, which we use to improve the model

• Need at least 2+ years of industry experience working deep in the weeds on hard ML problems.
• Negative example: just stringing together a bunch of pre-existing components together. Need signal that this person can think critically about different parts of the pipline
• Have a deep understanding of the “whole stack” when it comes to designing, training, evaluating and deploying machine learning models, especially large language models.
• Collected a new giant dataset
• Published research papers
• Played a critical role in shipping a new ML product that required custom components
• Writing distributed ML infrastructure
• Have debugged and fixed hard-to-find bugs in ML models
• Have a track record of successfully owning projects from start to finish.
• Have experience with generative models for various modalities.
• Experience working with proven tools: ML frameworks (Tensorflow, PyTorch, Jax, …), data processing frameworks (Spark, Beam, …).

Join our visionary team, on-site 3-5 days a week, at the forefront of next-generation AGI.