Skip to main content

​
Overview

Prizm’s AI-powered experiences leverage state-of-the-art language models to deliver intelligent, context-aware insights and recommendations that enhance user decision-making and automate complex workflows across the data platform.
Claude is the default AI model powering Prizm AI — including the Converse interface, metadata generation, and recommendation workflows. The platform also supports GPT and Gemini models, with configurable routing based on task complexity.

​
EPIC: AI-Driven Capabilities

This epic focuses on embedding AI-driven capabilities across the platform to enhance user productivity, automate complex tasks, and deliver intelligent, context-aware insights. It transforms how users interact with data by providing proactive recommendations, natural language interactions, and assisted decision-making.

​
Feature 1: GPT Prompts (PR-1767)

Prizm supports AI-driven prompt actions for metadata generation, recommendation, profiling, and quality workflows at multiple scopes.

​
Asset Description Prompts

Trigger AI to generate or update descriptions for data assets at any scope:
  • Update Asset Description for this asset
  • Update all Asset Descriptions
  • Update all Asset Descriptions for a Domain
  • Update all Asset Descriptions for a Product
  • Update all Asset Descriptions for an Application
  • Update all Asset Descriptions for a Tag

​
Attribute Description Prompts

Generate or update descriptions at the column/attribute level:
  • Update Attribute Description for this asset
  • Update Attribute Description for this attribute
  • Update all Attribute Descriptions (and scoped variants by Domain, Product, Application, Tag)

​
Term Recommendation Prompts

Recommend semantic business terms for classification:
  • Recommend Term for this Domain
  • Recommend Term for this Asset
  • Recommend Term for this Attribute
  • (and scoped variants by Product, Application, Tag)

​
Glossary Generation

  • Generate glossary — uses organization context (company name, summary) to create a full glossary of categories and related business terms
  • Generate glossary for a category
  • Generate glossary for a domain

​
Profiling & Quality Prompts

  • Recommend profile for this asset
  • Run profile for this asset
  • Recommend quality metric for this asset
  • Run quality metric for this asset

​
Acceptance Criteria

  • AI-generated recommendations only populate metadata when values are not already present, unless explicitly configured otherwise.
  • Existing approved or manually curated metadata is never overwritten automatically.
  • Generated recommendations move the entity into a “Ready for Review” workflow state.
  • Users can review, accept, or reject recommendations before final application.
  • Rejected recommendations are tracked and do not overwrite existing metadata.

​
Feature 2: Multi-Model Support (PR-1768)

Prizm expands beyond GPT-only models to support a configurable multi-model AI strategy.

​
Supported Models

ProviderModelsDefault?
Anthropic Claudeclaude-sonnet, claude-opus, claude-haiku✅ Default
OpenAI GPTgpt-4o, gpt-4, gpt-3.5Configurable
Google Geminigemini-pro, gemini-flashConfigurable

​
Model Routing by Task Complexity

Lightweight models (lower cost) for simpler operations:
  • Asset Description generation
  • Attribute Description generation
  • Metric Description, Purpose, Rationale
Thinking/advanced models for complex tasks:
  • Business Metric Recommendation
  • Governance Recommendation
  • Domain/Glossary Recommendation
  • Complex AI Insights
  • Advanced Quality Recommendations

​
Feature 3: Converse with MCP/Tools (PR-1769)

Prizm’s Converse interface delivers a Claude-style conversational AI experience powered by MCP (Model Context Protocol) integrations — no function writing required.

​
How It Works

Users interact with Prizm AI using natural language. The system automatically identifies and invokes the appropriate tools based on intent: 
User: "Recommend quality metrics for the orders table in the sales domain"
         │
         ▼
Converse → Intent Detection → Tool Selection → Execution → Recommendation Cards
         │
         ▼
User reviews and approves recommendations directly in the UI

​
Supported Governance Workflows

  • Domain recommendations
  • Glossary generation
  • Asset & Attribute description generation
  • Metric recommendations (profile, quality, business)
  • Profiling recommendations
  • Asset mapping recommendations

​
Converse Capabilities

The AI can:
  • Search available sources, assets, and governance structures
  • Analyze metadata and organizational context
  • Generate contextual recommendations with reasoning
  • Present actionable recommendation cards in the UI
Users can then:
  • Review individual or batch recommendations
  • Approve or reject recommendations
  • Trigger creation and execution actions directly from Converse

​
Contextual Awareness

Converse uses the following context to generate recommendations:
  • Existing domains and glossary
  • Asset metadata and profiling information
  • Organizational summaries
  • Lineage and usage insights
  • Historical quality patterns

​
Acceptance Criteria

Users interact with Prizm AI without invoking functions, APIs, tools, or code explicitly. The system autonomously identifies the correct MCP/tool based on intent.
The platform supports autonomous multi-step reasoning and execution workflows using connected tools and metadata context.
Users can trigger the following directly from Converse:
  • Create selected domains
  • Create glossary terms
  • Apply mappings
  • Run profile jobs
  • Run quality jobs
All AI-driven actions respect module licensing configuration, mode-level configuration, and user permissions and access controls.

​
AI Model Configuration

Model selection is configurable at the platform or workflow level by an Admin. See Model Configuration for setup instructions.