BI, Voir AI & ML tools

BI dashboards (Superset)

Open from Applications → BI (bi.<domain>). You sign in through SSO and must be a member of the workspace.

• Datasets: point them at gold tables in the lakehouse catalog, or at your DN Warehouse database for fast serving. Avoid bronze.
• Oracle sources are supported without an Oracle client:

  oracle+cx_oracle://<user>:<password>@<host>:1521/?service_name=<SERVICE>

• Governance. The pre-configured lakehouse (Trino) and warehouse connections have Impersonate logged-in user turned on, so queries run as the viewer and their row filters and masks apply.
  • ⚠️ Any database connection you add yourself runs as whatever user its URI names, which bypasses per-user rules. Keep Impersonate logged-in user ticked on any Trino connection you create, and don't put a privileged account in a shared connection.
  • Caches can serve one viewer's results to another. Ask your admin to keep caching off, or per-user, on datasets that carry row filters.
• Superset isn't connected to the semantic layer (cubes) yet. Build dashboards on gold tables.

Voir AI (licensed)

Voir gives you natural-language analytics over your data.

Page	Purpose
Chat	Ask questions
Twins (agents)	Configure an agent: integrations (lakehouse, semantic layer, documents), a data scope ("All accessible data" or "Selected only" with Catalog → Schema → Table), and skills
Skills	Reusable instructions or Q&A examples
Connectors	Register external MCP servers; Show available tools

Things to know:

• Voir's lakehouse connector queries through a read-only service account, not as the person chatting. Limit a twin's scope to the tables its audience may see.
• Answers are best when a twin uses cubes. Metrics are defined once, and the agent doesn't have to guess the joins.
• Empty Catalog dropdown or "Could not reach dn-lakehouse" is a platform configuration issue. Contact your admin.
• PII Guardrails (Data & AI Security) controls PII detection and redaction in AI answers.
• AI Gateway shows Not configured until your organisation's model endpoint is set up by an admin.

DocuContext (licensed)

A document knowledge base. You upload files into partitions and share them with members. Voir twins can use it as a source. Access is managed under Data & AI Security → DocuContext users.

ML & vector tools

These tools are available only from inside the platform (notebooks and workspace pods). They have no browser UI, by design.

Tool	Endpoint	Notes
MLflow	http://mlflow.mlops.svc.cluster.local:5000	Experiment tracking and model registry. Artifacts go through the server, so no S3 keys are needed.
Milvus	milvus.vector.svc.cluster.local:19530	Vector database, single node. REST v2 on the same port.
Redis Stack	redis-stack.cache.svc.cluster.local:6379	Cache / JSON / search. Password-protected; ask your admin.

All three are single-instance services. Treat them as shared team infrastructure: clean up experiments, and don't store the only copy of important data in Redis.