Skip to main content

AI-Assisted Capabilities

Qarion's AI-assisted features are designed to reduce manual governance work while keeping humans in control. Generated content is treated as a draft, recommendation, explanation, or triage signal that users can review before applying.

Discovery and Documentation​

AI Copilot​

The AI Copilot answers natural-language questions with catalog, lineage, quality, issue, and governance context. In tool-assisted mode, the Copilot can use approved Qarion API tools to search products, inspect product details, traverse lineage, and summarize quality status. Responses include sources so users can verify the underlying records.

Description Generation​

Product and field descriptions can be generated from metadata such as names, schema fields, data types, tags, ownership, environment, and lineage. Users review and edit generated descriptions before saving them to the catalog.

When embeddings are configured, catalog and documentation search can use semantic matching in addition to keyword search. This helps users find assets even when they describe the business concept rather than the exact table or column name.

Data Workflows​

AI Query Writer​

The Query Editor includes an AI Query Writer that can generate SQL from natural language, explain existing SQL, and refine a query with follow-up instructions. Generated and refined SQL is validated before insertion and is never executed automatically.

AI Classification​

AI classification suggests field-level sensitivity, PII, PHI, and tag classifications. Suggestions can combine pattern matching, lineage signals, and LLM analysis, then move through a review workflow before being applied.

Quality Check Suggestions​

AI quality suggestions recommend checks based on product schema, field types, metadata, and existing monitoring gaps. Stewards can review each suggestion before creating live quality checks.

AI Anomaly Explanation​

For anomaly or trend alerts, Qarion can gather context from sync events, upstream anomalies, recent quality failures, and schema drift, then generate a root-cause hypothesis with confidence and contributing factors.

Governance and Operations​

AI Impact Assessment​

AI impact assessment summarizes the likely risk, business impact, affected downstream products, contracts, owners, and recommended actions for product changes or issue investigations. Streaming variants show planner progress while the assessment is generated.

Ticket Triage and Tagging​

Issue tickets can receive AI-generated triage suggestions, including priority, root-cause category, assignee rationale, and tags. Users review suggestions before accepting them into the issue workflow.

AI Steward​

The AI Steward scans a space for governance work that needs attention, normalizes action items into an inbox, and can summarize the queue. It helps stewards focus on missing descriptions, stale metadata, classification work, and other maintenance tasks.

AI Change Log​

For AI products, Qarion can generate Annex IV-aligned change log entries from product snapshots, audit events, version milestones, and training run history. The generated markdown is a draft for model owners or compliance reviewers to edit and approve.

Standards Artifact Generation​

Architecture standards pages can produce AI-generated artifact proposals for review. Proposals are validated before users apply them as attachments or prepare repository rollout.

Administration and Controls​

Feedback Loop​

AI outputs can collect structured feedback such as thumbs up, thumbs down, accept, edit, and reject. Admin dashboards aggregate acceptance rates by feature so teams can identify weak prompts, missing context, or high-value workflows.

Logs, Usage, and AI Ops​

AI interactions are logged with feature, model, status, token usage, guardrail metadata, and workflow links where available. Admin AI Ops pages surface usage, failures, guardrail impact, workflow runs, and evaluation status.

Guardrails and Budgets​

Qarion supports product-scoped guardrails, runtime guardrails, feature gates, token reservations, daily limits, and admin budget controls. These controls keep AI features observable, permission-aware, and aligned with tenant policies.

Provider and Embedding Configuration​

Superadmins configure LLM providers, model routing, and embedding providers from the admin area. Embeddings can power semantic search for catalog and documentation content, while LLM configuration powers generation, triage, and Copilot workflows.