A New Approch to Model Orchestration
Built around objects, resources, and jobs — trac offers a new, universal, and unopinionated framework for structural analytics
BOOK A DEMO
Business Objects
Whatever the use case, platform users primarily create and interact with four trac object types:
Data object: Refers to and describes a collection of data records in trac's primary storage, which can be used in calculations
Model object: Refers to and describes a discrete unit of code in a repository trac can acccess, which can be used in calculations
Flow object: The blueprint for a complex calculation composed of multiple models arranged into an execution graph
Job object: An instruction for trac to orchestrate a calculation — such as running a specific model or flow
System Resources
When orchestrating workflows trac will use the capabilities of your existing technologies. A typical deployment will make use of five system resources:
Internal storage: Storage location(s) which hold the platform’s primary data — locations to which trac alone has write access
External storage: Storage location(s) which trac does not control but can access for importing and exporting data
Model repository: One or more locations within a version control system (e.g. Git) where models are stored
Execution service: The compute capabilities of your underlying infrastructure which trac uses to execute calculations.
Runtime environment: The set of libraries and dependencies that are made available to models at runtime
Jobs and Actions
Any analytic workflow can be supported on trac — by appluing the following actions to your models and data:
1. Import model: Bring in a new model or model version from a repository
2. Upload/import data: Bring in data via the UI or a batch process
3. Build flow: Combine multiple models into an execution graph
4. Run job: Execute a model or flow against a set of data inputs
5. Download/export data: Extract data via the UI or a batch process
