DAFT
Daft is an open-source, high-performance data processing engine optimized for handling multimodal data—including structured tables, text, images, videos, and tensors—in AI and machine learning workflows.[1]
Developed by Eventual, Inc., it features a Rust-based core for efficiency, paired with a Python-native API that supports declarative queries and integrates natively with libraries like PyTorch, NumPy, Pandas, and Hugging Face, while scaling from local execution to petabyte-level distributed clusters without requiring JVM overhead.[1][2]
The engine emerged from data processing challenges encountered by its founders during autonomous vehicle projects at Lyft, addressing limitations in existing tools for large-scale, unstructured data handling.[3]
Eventual raised $30 million in funding across seed and Series A rounds in 2025, led by investors including Felicis and CRV, to expand Daft's capabilities and launch Eventual Cloud for enterprise multimodal AI infrastructure.[4]
Daft has gained adoption among prominent users such as Amazon, where it delivered 24% efficiency gains and saved equivalent to 40,000 years of EC2 vCPU annually, alongside Essential AI, Together AI, ByteDance, and Cloud Kitchens for mission-critical pipelines.[1][5]
Key characteristics include zero-copy user-defined functions via Apache Arrow for reduced memory usage, universal connectivity to storage like S3 and Delta Lake, and intelligent memory management for reliability in AI model training and inference.[1]