7. Foundry IQ

Introduction to Foundry IQ

Foundry IQ combines agentic retrieval with intelligent ranking. Instead of simple keyword search, agents autonomously decide which knowledge sources to query and how to rank results.

Agentic Retrieval

Unlike static RAG, agentic retrieval lets the agent:

Choose which knowledge base to query based on the question
Rewrite queries for better retrieval
Combine results from multiple sources
Decide when to fall back to model knowledge

Ranking Techniques

Technique	Description
Semantic Ranking	Re-rank results by semantic similarity to the query
Hybrid Ranking	Combine vector similarity + keyword BM25 scores
LLM Re-ranking	Use an LLM to judge relevance of each result
Reciprocal Rank Fusion	Merge ranked lists from multiple retrieval sources

Knowledge Base Setup

from azure.ai.projects import AIProjectClient

# Create a knowledge base connection
kb = client.knowledge_bases.create(
    name="company-docs",
    description="Internal company documentation",
    storage_connection_string=conn_str,
    search_endpoint=search_endpoint,
    chunk_size=512,
    overlap=64
)

# Attach to agent
agent = client.agents.create_agent(
    model="gpt-4o",
    name="IQAgent",
    knowledge_bases=[kb.id]
)

Creating Agents via Portal & Python

You can create IQ agents either through the Foundry portal (no-code) or programmatically via the SDK (as shown above).

✏️ Exercise: Create a knowledge base from a set of internal documents. Build an agent with Foundry IQ that can answer questions across multiple knowledge bases. Test queries that require combining information from two different sources.