Keep pulling the thread on Nico West.
Rerun.ai made its visualization tool open source because it believes visualization is too critical and widely used across different roles to be effectively monetized with a closed-source, per-seat license model.
Building high-performance and flexible data pipelines is easier for LLM teams because they can leverage mature storage formats like Parquet and Iceberg and data processing engines like Spark, whereas equivalent tools for multimodal robotics data do not exist.
Rerun.ai has redesigned its core data model approximately four times to better support its users.
Rerun.ai provides a logging product and a system of record for companies in robotics, embodied AI, and augmented reality.
When Weights & Biases first started working with OpenAI, OpenAI's primary focus was on robotics, prior to its work on GPT.
Rerun.ai's platform is used by hedge funds and is now heavily adopted in the "learning first" robotics sector.
Rerun.ai is building a commercial data platform, described as a cloud database or data lakehouse, to manage data pipelines for robotics training.
Rerun.ai's commercial platform focuses on the data pipelines leading up to model training, differentiating it from companies like Weights & Biases which focus on the training process itself.
Rerun.ai built a custom in-memory database from scratch to power its high-performance visualization application.
Rerun.ai designed its data model based on the entity component systems (ECS) paradigm commonly used in video game development.
The "ChatGPT moment" demonstrated the power of scalable machine learning, which primed the industry to seek similar scalable methods for robotics.
The availability of open source models, heavily driven by Hugging Face, has enabled more practical and scrappy robotics companies to build working products.