Intelligence · Module
Data Science Workbench
Governed notebooks and models, deployed as skills only after evaluation and approval
Data Science Workbench is planned to give data scientists managed notebooks that see data only through permitted channels, with jobs, feature sets and a model registry with model cards. Models can be deployed as scoring skills after an evaluation gate and approval.
Data science inside the guardrails
Notebooks often become the place where governance ends. Data Science Workbench, planned for W4 and requiring Data Fabric, keeps them inside it.
Governed access
Notebooks run with the user's tenant identity. Data access goes only through the semantic layer, datasets and APIs, under the user's permissions and Data Catalog policies. No outbound network is allowed by default, packages come from an allow-list and secrets come from the platform secret manager.
Jobs and features
Jobs run notebooks and scripts on a schedule or a trigger, with compute quotas and a cost view. Feature sets carry definitions, owners, freshness and lineage to datasets. Notebooks and pipelines sync with Git. GPU compute through One3 Compute is planned for W5.
Models with accountability
The model registry stores every model with a model card: data window, features, metrics, intended use and limitations. AutoML for tabular classification and regression includes leakage checks. A notebook copilot suggests code, explains results and writes queries against the semantic layer; code runs only when the user runs it.
From model to skill
A model can be deployed as a scoring skill or an Action-callable tool for agents, but only after an evaluation gate and an R3 approval, with versioned rollback. Deployed models are monitored with an inference log (inputs redacted by classification), drift, latency, error-rate and label-performance monitors. A breach raises a signal and a rollback proposal.
How it fits
The Workbench sits on top of Data Fabric datasets and BI Studio metrics, so a data scientist starts from the same certified definitions the rest of the business uses. Models it produces can feed Forecast & Simulation, score records for Digital Employees, or support Decision Engine options. Compute and notebook engines such as JupyterLab and MLflow are named as build choices and still need their licence and version checks before production.
Features
What Data Science Workbench includes
Managed notebooks
Tenant identity; data only through semantic layer, datasets and APIs under user permissions.
No outbound network by default
Package allow-list and platform-managed secrets.
Jobs
Scheduled or triggered runs with compute quotas and cost view.
Model registry with model cards
Data window, features, metrics, intended use and limitations.
Notebook copilot
Code suggestions, result explanations and semantic-layer queries.
Feature sets
Definitions, owners, freshness and lineage.
AutoML with leakage checks
Tabular classification and regression.
Deploy as a skill
Scoring skill or Action-callable tool after evaluation and R3 approval, with rollback.
Git sync
For notebooks and pipelines.
Model monitoring
Inference log, drift, latency and error monitors with rollback proposals.
Workflow
Notebook to deployed skill
- 1
Open a notebook with governed data access
- 2
Build features and train a model
- 3
Register the model with a model card
- 4
Run the evaluation gate
- 5
An approver authorises deployment as a skill
- 6
Monitors watch drift and errors and propose rollback on breach
AI
AI skills and Digital Employees
Notebook copilot
Suggests code and explains results; the user runs the code.
AutoML
Produces candidate models with leakage checks.
Model card drafting
Drafts cards for the owner to complete.
Digital Employees that use this module
They act through the same governed Actions as people; risky steps wait for approval.
Frequently asked questions
Can a notebook send data outside?
Not by default. Outbound network is blocked and packages come from an allow-list.
Can agents use my models?
After an evaluation gate and approval, a model can be deployed as a scoring skill with versioned rollback.
When is it planned?
W4, with GPU compute planned for W5.
See Data Science Workbench in your workflow
Tell us about your team and the outcomes you want. We will walk you through the apps, the Digital Employees and the plan for your industry.