What is Feathur?
Feathur is an open-source feature store designed to help teams effectively manage machine learning features across the full model lifecycle. As models move into production, Feathur enables you to log, store, and retrieve large volumes of features for model training and inference.
Key capabilities include:
- Scalable storage for features using Redis or PostgreSQL as a backing database
- APIs to read and write features with low latency
- Transformation and processing of raw data into features
- Integration with popular ML frameworks like PyTorch, TensorFlow, and Spark
- Caching for fast retrieval of frequently used features
- Built-in monitoring and telemetry
- Query optimizations for retrieving related sets of features
With Feathur, teams can simplify feature pipelines, reduce duplicate work across projects, provide a centralized view of the features used across the organization, and increase the velocity, quality and reproducibility of models. The open-source model makes Feathur easily integrable into an existing tech stack.