> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dqlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI & ML Capabilities

> AI/ML capabilities, availability, and trust overview for Prizm.

<script type="application/ld+json">
  {`{
            "@context": "https://schema.org",
            "@type": "TechArticle",
            "headline": "AI & ML Capabilities",
            "description": "AI/ML capabilities, availability, and trust overview for Prizm.",
            "url": "https://docs.dqlabs.ai/platform/ai-and-ml-capabilities",
            "publisher": {
              "@type": "Organization",
              "name": "DQLabs Inc",
              "logo": "https://media.brand.dev/332adc35-5bc4-4d2b-bf78-256aa4a5e414.svg"
            }
            }`}
</script>

*As of July 23, 2026.*

## Platform Architecture

Prizm is an "AI-native" control plane — AI is embedded in how context is gathered, how signals are interpreted, how priorities are set, and how actions are orchestrated, rather than bolted on as a chat feature.

## 1. AI/ML-Powered Capabilities, Descriptions, Use Cases & Availability

DQLabs does not publish a formal GA/Preview/Roadmap matrix. Status below is inferred from how each capability is presented on DQLabs' current product pages: features described as live, core platform functionality are marked GA; the one feature explicitly gated behind a public waitlist is marked Private Beta / Roadmap.

| Capability                                                                                                                                                                            | Description & Business Use Case                                                                                                                                                                                                       |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **AI/ML-Powered Anomaly & Drift Detection**                                                                                                                                           | Continuously profiles data and pipelines and applies statistical/ML models to flag freshness, volume, schema, and distribution drift in real time, instead of relying only on static thresholds.                                      |
| ***Use case:***\_ Catches silent pipeline breaks and data corruption before they reach dashboards, reports, or downstream ML models — reducing analytical errors and data downtime.\_ |                                                                                                                                                                                                                                       |
| **Automated Data Profiling**                                                                                                                                                          | Auto-discovers structure, patterns, and statistical properties of datasets on a continuous or scheduled basis without hand-written rules.                                                                                             |
| ***Use case:***\_ Speeds up onboarding of new data sources and surfaces quality gaps (nulls, outliers, format drift) without dedicated engineering time.\_                            |                                                                                                                                                                                                                                       |
| **Alert Clustering & Prioritization**                                                                                                                                                 | Groups related signals into a single incident and ranks them using lineage, usage/data-rate change, and business criticality rather than surfacing every alert individually.                                                          |
| ***Use case:***\_ Reduces alert fatigue so data teams focus on the handful of incidents that actually affect revenue-critical dashboards or models.\_                                 |                                                                                                                                                                                                                                       |
| **Autonomous Issue Resolution**                                                                                                                                                       | Recommends (re-run job, default value, quarantine bad records), validated in staging first.                                                                                                                                           |
| ***Use case:***\_ Cuts mean-time-to-resolution for well-understood, recurring failure patterns.\_                                                                                     |                                                                                                                                                                                                                                       |
| **Specialized Role-Driven AI Agents**                                                                                                                                                 | Dedicated agents for Quality, Observability, Cataloging, and Governance collaborate over a shared context layer instead of one generic bot handling everything.                                                                       |
| ***Use case:***\_ Lets an enterprise dial in automation per domain — e.g., full automation for routine quality checks, human sign-off for governance actions.\_                       |                                                                                                                                                                                                                                       |
| **Intelligent Cataloging & Criticality Scoring**                                                                                                                                      | Automates data-asset discovery/classification and scores assets by business criticality using usage and lineage signals.                                                                                                              |
| ***Use case:***\_ Gives data leaders a prioritized inventory of what data matters most for governance and compliance scoping.\_                                                       |                                                                                                                                                                                                                                       |
| **Semantic Intelligence / Semantic Layer**                                                                                                                                            | Maps technical fields to business terms and identifies meaningful relationships between entities across disparate source systems.                                                                                                     |
| ***Use case:***\_ Makes data self-service-friendly for business users and grounds GenAI/RAG applications in validated business meaning.\_                                             |                                                                                                                                                                                                                                       |
| **AI-Assisted Stewardship Modes**                                                                                                                                                     | Routes every detected action into one of three modes — Autonomous, Human/AI Assisted, or Action Needed — based on confidence and risk.                                                                                                |
| ***Use case:***\_ Gives regulated industries a governance dial: automate low-risk/high-confidence actions while keeping humans in the loop for high-risk ones.\_                      |                                                                                                                                                                                                                                       |
| **Interactive Lineage Explorer**                                                                                                                                                      | Animated, auto-traced graph of schema, volume, and freshness changes with upstream/downstream dependency mapping.                                                                                                                     |
| ***Use case:***\_ Cuts root-cause-analysis time when investigating an incident.\_                                                                                                     |                                                                                                                                                                                                                                       |
| **Business Impact Visualizer**                                                                                                                                                        | Maps a data issue to the downstream business KPIs and dashboards it affects and visualizes how it propagates.                                                                                                                         |
| ***Use case:***\_ Helps leaders and stewards judge business urgency, not just technical severity.\_                                                                                   |                                                                                                                                                                                                                                       |
| **GenAI-Enabled Remediation & Rule Generation**                                                                                                                                       | Uses generative AI, combined with the semantic layer, to draft/recommend data-quality rules and plain-language explanations of anomalies and fixes.                                                                                   |
| ***Use case:***\_ Reduces the no-code effort stewards need to write new quality checks.\_                                                                                             |                                                                                                                                                                                                                                       |
| **Multi-Agent Agentic AI Data Management** (unstructured data, RAG grounding, self-healing pipelines)                                                                                 | Extends agent orchestration to unstructured data validation, automated RCA, and self-healing pipelines. ***Use case:***\_ Targets organizations building RAG/GenAI applications that need validated, AI-ready unstructured content.\_ |

## 2. Underlying AI/ML Technology

The following technology layers are enabled by OpenAI and Claude-supported models. Gemini support is not yet generally available:

* Statistical/ML anomaly and drift detection with configurable thresholds — the platform FAQ confirms anomaly detection is "not limited to the AI & ML-based default configuration"; users can set manual thresholds alongside the ML defaults.
* A semantic/context layer that uses AI to connect technical metadata to business terms and to find relationships between data entities.
* Generative AI ("GenAI-enabled remediation") for rule generation and remediation suggestions, layered on top of the semantic engine.
* Role-driven, multi-agent orchestration (an "agentic AI" design) coordinating Quality, Observability, Cataloging, and Governance agents around a shared context store.

### Per-capability model details

| Capability                                              | AI/ML technique                                                                                                                                                                                  | Model / provider                                                                                                                                                                                                                                              |
| ------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Alert Drift / Threshold Detection (UnivariateModel)** | Classical statistical forecasting — not an LLM. Model is selected by rule based on data length, stddev, and seasonality.                                                                         | EWMA (t-distribution CI), AutoReg, Facebook Prophet, SARIMAX/ARIMA (via AutoARIMAForecaster with auto seasonality detection), Hampel filter for outlier handling. Open-source libraries: statsmodels, Prophet, scipy. Runs in-process — no external API call. |
| **AI SQL Assistant**                                    | LLM                                                                                                                                                                                              | GPT-5.1 nano, served via **Azure OpenAI Service** (not OpenAI's public API).                                                                                                                                                                                  |
| **Description Auto-Completion**                         | LLM                                                                                                                                                                                              | GPT-5.1 nano, served via **Azure OpenAI Service**.                                                                                                                                                                                                            |
| **Profile Recommendation Agent**                        | Two components: (1) deterministic rules/heuristics for scheduling and scope decisions — no AI; (2) a single LLM call per table to recommend the best timestamp column for incremental profiling. | LLM component: likely GPT-5.1 nano family via internal `create_llm_instance` helper — confirm exact model/deployment with the platform team.                                                                                                                  |

## 3. Data Processing Requirements & Model Training on Customer Data

Prizm uses your metadata to enable data observability, data quality, and context-aware operations, without retaining any customer data from your data platforms. The core architecture is built to operate primarily on metadata, lineage, and statistical profiles rather than always requiring full raw-data movement out of the customer's environment.

The platform supports both on-premises and cloud/hybrid deployments and is suited for multi-cloud workloads — implying data processing location is configurable to the customer's environment rather than fixed to a single DQLabs-hosted store.

No Customer Data is also used for any model training. Currently no custom models are used and only OpenAI/Claude models are supported using Azure / AWS hosting environments in a secure, private network.

## 4. Hosting & Deployment Details

|                                |                                                                                                                                                                                                                                                                                               |
| ------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Deployment models**          | On-premises, private/hybrid cloud, and public cloud — DQLabs supports multi-cloud and hybrid-cloud workloads.                                                                                                                                                                                 |
| **Marketplace availability**   | Listed on AWS Marketplace; technology-partner badges/integrations exist for Snowflake, Databricks, Azure, and Google Cloud Platform.                                                                                                                                                          |
| **Data platform integrations** | Snowflake, Databricks, BigQuery, Redshift, Azure Synapse, SAP HANA, Oracle, PostgreSQL, IBM DB2, S3/ADLS, and orchestration tools (Airflow, ADF).                                                                                                                                             |
| **AI/ML compute location**     | U.S East/West for USA region, Ontario for CA region, Dubai for Middle East region, Germany for European region, Singapore for APAC region. Also note that ML inference and any LLM calls can run inside the customer's deployment boundary (on-prem/VPC) apart from a DQLabs-managed service. |

## 5. Security, Privacy & Compliance Controls

|                            |                                                                                                                                                                                                                                                                                     |
| -------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Compliance**             | SOC 2, GDPR, and HIPAA                                                                                                                                                                                                                                                              |
| **Access control**         | Role-Based Access Control (RBAC) with predefined roles (Admin, Owner, User) and granular, feature-level permissions, per the platform FAQ.                                                                                                                                          |
| **Identity integrations**  | Azure Active Directory, Okta, PingFederate, and SAML integrations with SSO/federated identity support.                                                                                                                                                                              |
| **Regulatory positioning** | SOC 2, GDPR, and HIPAA                                                                                                                                                                                                                                                              |
| **Website privacy policy** | [https://www.dqlabs.ai/privacy-and-legal/dqlabs-end-user-license-and-services-agreement/](https://www.dqlabs.ai/privacy-and-legal/dqlabs-end-user-license-and-services-agreement/) · [https://www.dqlabs.ai/website-privacy-policy/](https://www.dqlabs.ai/website-privacy-policy/) |

## 6. Prerequisites, Licensing & Configuration

### Licensing model

Prizm is sold as a single unified package plus optional add-ons (no tiered plans), custom-quoted to the customer's environment:

* Included in the base package: full Prizm platform (Data Quality, Observability, Context); one workflow integration (ServiceNow or Jira); one data catalog integration; one data source type with unlimited assets, data volume, users, and DQ rules; two alert channels (email plus Teams/Slack/chat); and 8×5 onboarding, training, and support.
* Available add-ons: native cataloging; additional data source types (warehouse, lake, pipeline, or BI — each unlimited once added); additional app integrations; additional tenants (for divisions, regions, or pre-production); a non-production sandbox; orchestration compute for tighter SLAs; extended support (12×5 or 24×7, or hourly experts); and custom development.
* No per-seat, per-row, or per-asset fees within a connected data source — pricing scales by the number of source connectors and add-ons, not by users or data volume.

### Prerequisites / configuration steps

* Connect at least one supported data source (warehouse, lake/lakehouse, pipeline tool, or BI tool).
* Optionally connect an existing data catalog (e.g., Collibra, Alation) or add native cataloging as an add-on.
* Configure identity/SSO (Azure AD, Okta, PingFederate, or SAML) and assign RBAC roles (Admin, Owner, User) before granting user access.
* Set up alert-channel integrations (email plus Slack/Teams/chat) and, if used, a workflow tool (Jira or ServiceNow) for issue routing.
* Pricing and final scope are only available via a custom quote ("Talk to Sales") — there is no public self-serve price list.

### Per-capability prerequisites and configuration

| Capability                            | Prerequisites & licensing                                                                                                                                                                                                                         | Configuration steps                                                                                                                        |
| ------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ |
| **Alert Drift / Threshold Detection** | No external vendor licensing. Uses open-source libraries (statsmodels, Prophet, scipy) included in the Prizm deployment.                                                                                                                          | No customer-facing configuration required. Runs automatically as part of metric collection and alerting.                                   |
| **AI SQL Assistant**                  | Azure subscription with OpenAI Service enabled; GPT-5.1 nano model deployment provisioned in-tenant; API key or Azure AD-based authentication configured in Prizm.                                                                                | Confirm Azure region and tenant with the platform/infra team. Verify Azure AD auth or API key is set in Prizm's LLM configuration.         |
| **Description Auto-Completion**       | Same Azure OpenAI prerequisites as the AI SQL Assistant (shared GPT-5.1 nano deployment).                                                                                                                                                         | No additional configuration beyond the shared Azure OpenAI deployment.                                                                     |
| **Profile Recommendation Agent**      | LLM provider access via `create_llm_instance` (same Azure OpenAI deployment — confirm with platform team). MCP tool integration per source type (Snowflake, Databricks, etc.) must be configured to enable the `compute_timestamp_features` step. | Verify `create_llm_instance` points to the correct model deployment. Confirm MCP tool contracts are active for each connected source type. |