ML Studio
ML Studio is where you extend Athena's models with your own. ConverSight ships with production-ready forecasting and planning models that cover most supply chain workflows out of the box — but every organization has the workflows that need their own logic. ML Studio is where you build, train, version, and deploy the custom models that handle those organization-specific cases, and where those models become available to the rest of the platform.
This is primarily a Data Engineer surface. Custom models built here feed Athena's Demand Planner and Supply Planner agents, power the forecasts that show up in Storyboards, and back the scheduled outputs of AI Briefing.
What ML Studio Provides
| Capability | What it does |
|---|---|
| Model authoring | Build models in the supported frameworks against the same governed semantic layer the rest of the platform uses. No separate data pipeline required. |
| Training and validation | Train against historical data with automatic train/validation/test splits and accuracy tracking against the metrics that matter for your use case. |
| Versioning | Every model deploy is versioned. Roll back to a prior model if a new one underperforms, with full audit history of which version was live when. |
| Deployment | Promote models into the agent runtime that Athena uses. No separate model-serving infrastructure to stand up. |
| Monitoring | Track model performance drift over time. Athena flags when a model's accuracy starts diverging from its training baseline, so retraining happens on signal rather than schedule. |
Where Custom Models Fit
The pre-built models that ship with ConverSight cover the common patterns: demand forecasting, replenishment optimization, supplier risk scoring, anomaly detection. Custom models in ML Studio are for the cases those don't cover:
- Organization-specific forecasting — a unique product hierarchy, an unusual sales motion, an industry-specific seasonality model.
- Domain-specific scoring — your own definition of supplier risk, your own customer-segmentation logic, your own margin-leak detection.
- Specialty workflows — pricing optimization, route planning, network design, anything your operations team has historically managed in spreadsheets or custom code.
A custom model built here doesn't replace Athena's pre-built capability — it extends it. The Demand Planner agent will use the right model for the right SKU class, choosing between built-in and custom by configuration rather than re-architecture.
Coming next
Detailed walkthroughs for model authoring, training cycles, deployment, and monitoring are in progress. In the meantime: see the Data Engineer persona page for how ML Studio fits into the broader data-engineering workflow, and the Workbench → Data Studio section for the data layer that ML Studio models build on.