AI & Machine Learning Types
AI and ML types are designed for managing the full machine learning lifecycle — from model development through deployment and monitoring. These types unlock a specialized Model Details tab with AI-specific metadata.
For repository-style management of AI and ML products, including organization-wide discovery, versions, artifacts, chunked uploads, checksums, and current artifact selection, see the Model Repository.
Available Types
ML Model
The primary type for machine learning models.
- Icon: Brain · Color:
#f59e0b(amber) - Created by: Manual registration or ML pipeline integration
AI System
An AI system registered for regulatory compliance and governance.
- Icon: Brain · Color:
#f59e0b - Created by: Manual registration
AI Systems are designed for EU AI Act compliance, with built-in risk classification.
LLM Agent
A Large Language Model agent, chatbot, or conversational AI system.
- Icon: Brain · Color:
#f59e0b - Created by: Manual registration
AI App
An AI-powered application or service.
- Icon: Brain · Color:
#f59e0b - Created by: Manual registration
Agent
An LLM-powered autonomous agent capable of tool-calling, multi-step reasoning, and agentic workflows.
- Icon: Bot · Color:
#6366f1(indigo) - Created by: Manual registration
Model Details Tab
All AI/ML types unlock the Model Details tab, which replaces the standard Schema tab. This tab captures:
Core Properties
| Property | Description | Examples |
|---|---|---|
| Architecture | Model family or architecture | Transformer, XGBoost, CNN, Random Forest |
| Framework | Training framework | PyTorch, scikit-learn, TensorFlow, JAX |
| Intended Use | What the model is designed for | Churn prediction, fraud detection, content recommendation |
| Limitations | Known failure modes and constraints | Low precision on edge cases, works only for English text |
Prediction Semantics
| Property | Description | Examples |
|---|---|---|
| Prediction Entity | The entity type being predicted on | customer_id, transaction_id, document_id |
| Model Objective | Target prediction goal | Churn probability, lifetime value, sentiment score |
Deployment & Lifecycle
| Property | Description | Values |
|---|---|---|
| Deployment Status | Current lifecycle stage | Development, Staging, Production, Retired |
| Lifecycle Stage | Broader AI lifecycle tracking | Ideation → Design → Training → Validation → Deployment → Monitoring → Retirement |
Metrics & Hyperparameters
- Evaluation Metrics — Named metrics tracked per model version (e.g., AUC-ROC, F1, precision, recall)
- Training Metrics — Metrics from the training process (e.g., loss, val_loss, accuracy)
- Hyperparameters — Flexible JSON storage for training configuration (learning rate, epochs, batch size, etc.)
Each model version can record specific metric values, enabling performance tracking across releases.
AI Governance Features
AI/ML types integrate with Qarion's governance framework:
- Risk Classification — Aligned with EU AI Act categories: Unacceptable, High, Limited, Minimal
- Risk Assessments — Configurable questionnaires for compliance evaluation
- Impact Analysis — Lineage-based downstream impact when models change
- Change Requests — Formal approval workflows for model updates
Tab Differences vs. Standard Types
| Feature | Available? | Notes |
|---|---|---|
| Overview | ✅ | Standard overview with markdown documentation |
| Schema | ❌ | Replaced by Model Details tab |
| Model Details | ✅ | Architecture, framework, metrics, hyperparameters |
| Data Profiling | ❌ | No source table to profile |
| Env Diff | ✅ | Compare across environments |
| Quality Health | ✅ | Standard quality checks |
| Lineage | ✅ | Training data → model → downstream consumers |
| Governance | ✅ | With AI-specific risk classification |
| Versions | ✅ | Track model releases with metric snapshots |
Technical Sidebar
The metadata sidebar shows ML-specific information:
- Architecture and Framework badges
- Deployment Status indicator
- Prediction Entity and Model Objective
- Standard governance assignments (Owner, Steward)