Quick Overview of Foundry IQ #ai #microsoftfoundry

Key Concepts

• Foundry IQ: A new capability built on Azure AI Search enabling the creation of knowledge bases for AI applications and agents.
• Knowledge Base: A collection of one or more knowledge sources used by Foundry IQ.
• Knowledge Sources: Data repositories that Foundry IQ can query, including Azure AI Search indexes, Fabric/Fabric IQ SharePoint sites, remote SharePoint (via Work IQ), the web, and MCP servers.
• Reasoning Effort (Low/Medium): Controls the complexity of Foundry IQ’s search and analysis process. Higher effort allows for self-reflection and iterative refinement.
• Extractive Data: The raw information retrieved from knowledge sources, intended for use by agents or applications.
• Answer Synthesis: Foundry IQ’s ability to generate a natural language answer directly from the information found in knowledge sources.

Foundry IQ: Enabling Knowledge Bases for AI Applications

Foundry IQ represents an advancement built upon the foundation of Azure AI Search, specifically designed to facilitate the creation of robust knowledge bases. These knowledge bases are not limited to a single data source; instead, they are constructed from one or more knowledge sources. These sources demonstrate significant flexibility, extending beyond traditional Azure AI Search indexes.

Supported Knowledge Sources

Foundry IQ supports a diverse range of data sources. These include:

• Azure AI Search Indexes: Existing and standard Azure AI Search indexes can be directly integrated as knowledge sources.
• Fabric/Fabric IQ SharePoint Sites: Data residing within Microsoft Fabric and Fabric IQ-specific SharePoint sites can be indexed and utilized.
• Remote SharePoint (via Work IQ): Foundry IQ can query SharePoint sites remotely using Work IQ’s native semantic index, eliminating the need for data duplication.
• The Web: Foundry IQ possesses the capability to query information directly from the internet.
• MCP Servers: Integration with MCP (Microsoft Content Platform) servers is also supported, allowing access to data stored within these systems.

Reasoning Effort and Iterative Refinement

A key differentiator of Foundry IQ is its ability to intelligently manage the search process based on a defined reasoning effort. This parameter dictates the complexity of the analysis performed.

• Low/Medium Reasoning Effort: At these levels, Foundry IQ dynamically creates a search plan, selecting only the most relevant knowledge sources for the given query. This optimizes efficiency by avoiding unnecessary data processing.
• Medium Reasoning Effort – Self-Reflection: With medium reasoning effort, Foundry IQ goes a step further. It performs a self-assessment to determine if the initial answer is optimal. If not, it initiates another iteration of the search and analysis process, refining the results.

Output Options: Extractive Data vs. Answer Synthesis

Foundry IQ offers two primary output options, catering to different application needs:

• Extractive Data: This option provides the raw, uninterpreted information retrieved from the knowledge sources. This is ideal for scenarios where an AI agent or application requires granular control over the data and needs to perform its own analysis and grounding.
• Answer Synthesis: Foundry IQ can directly generate a natural language answer based on the prompt and the information gathered from the knowledge sources. This is suitable for applications requiring a concise and readily understandable response.

Logical Connections & Synthesis

The core functionality of Foundry IQ lies in its ability to intelligently connect to diverse data sources, apply varying levels of reasoning to refine search results, and deliver information in a format tailored to the specific needs of the consuming application. The system’s ability to self-reflect and iterate on its search process, particularly at the medium reasoning effort level, represents a significant advancement in knowledge retrieval for AI.

As stated, Foundry IQ is a “fantastic way to now get knowledge for my AI apps and agents,” highlighting its practical value in empowering AI-driven solutions with access to comprehensive and relevant information.