Welcome to the Eventual blog
Join us as we explore innovative ways to handle multimodal datasets, optimize performance, and simplify your data workflows.
Finding a Needle in the Haystack: Querying Physical AI Data with Daft
Pose + semantic search over Apple's EgoDex hand-manipulation dataset with Daft: SigLIP embeddings meet hand-pose geometry. Ctrl+F for physical AI data.
How we made our LeRobot video reader up to 15× faster
LeRobot has emerged as the dominant open format for robot learning data, but decoding frames is expensive. Here's how we made Daft's native LeRobot reader up to 15× faster.
Daft v0.7.17: LeRobot Datasets, HDF5 Files, and Local LLM Inference
Daft v0.7.17 ships daft.datasets.lerobot for LeRobot v3 robotics data, native HDF5 file support, and local LLM inference via the Transformers provider.
Daft v0.7.16: DROID Robotics Dataset, PyTorch DataLoader, and Resilient File Reads
Daft v0.7.16 ships DROID robotics dataset support, a native PyTorch DataLoader, daft.concat() for multi-DataFrame workflows, and ignore_corrupt_files for resilient batch processing.
Finding a Needle in the Haystack: Querying Physical AI Data with Daft
Pose + semantic search over Apple's EgoDex hand-manipulation dataset with Daft: SigLIP embeddings meet hand-pose geometry. Ctrl+F for physical AI data.
Daft v0.7.15: Safe Type Conversions, Flight Shuffle Optimizations, and PostgreSQL Support
Daft v0.7.15 ships with try_cast for safe type conversion, Flight shuffle LZ4 compression, UUIDv7 timestamp extraction, and PostgreSQL support.
Disk is the data plane: Flight Shuffle in Daft
How Daft rebuilt distributed shuffle around Arrow Flight, local disk, and streaming reads to handle multi-terabyte workloads.
The limiting factor in physical AI isn't compute or architecture - it's data
Robotics is hitting a data wall. The architecture debate gets much of the attention, but the data constraint is more fundamental.
First-class observability in Daft
A new dashboard, per-operator memory attribution, and OTel endpoints for your existing collector. Everything you need to see what Daft is doing with your query.
Daft v0.7.14: Parquet Reader Rewrite, Streaming Distributed Limits, and UUIDv7
Daft v0.7.14 rewrites the Parquet reader on arrow-rs for up to 17x faster remote reads, ships streaming distributed limits, and adds native UUIDv7 generation.