SKIL bridges the gap between data scientists and deployment (devops) engineers by providing all of the necessary tools to build, train, and deploy a model.
Because deploying a model requires input from more than just data scientists, SKIL has a collaborative UI and extensive Command Line (CLI) to help devops engineers and product managers participate in fine-tuning and serving models at scale. SKIL reduces friction between all parties in the data science workflow and helps you scale your model faster.
Teams using SKIL can expect support for the following workflows:
- Model and data configuration
- DNN Training
- Collaborative user interfaces for data and results
- Versioning of experiments and models
- Scalable microservice deployment architecture
- APIs for model serving
- Management UI
To learn more and get started with the SKIL workflow, read the Workflow Overview.
Using well-known tools for data science and distributed systems, SKIL is built on the JVM and uses several bindings to unite common deep learning frameworks and distributed storage systems.
SKIL is compatible with the Apache Spark and Hadoop ecosystem, allowing for more complex storage of data and operations such as Batch Inference.
SKIL itself is accessible as a microservice in your architecture, and exposes key APIs for serving models and transforming data. See our API Reference.