5. End-to-End RAG

RAG Basics

Retrieval-Augmented Generation (RAG) combines information retrieval with LLM generation. The model queries a knowledge base, retrieves relevant chunks, and generates answers grounded in those chunks.

Vector Embeddings

Documents are split into chunks, each chunk is converted to a vector embedding using an embedding model, and stored in a vector index.

from azure.ai.inference import EmbeddingsClient

embeddings = EmbeddingsClient(endpoint="...", credential=cred)
response = embeddings.embed(
    model="text-embedding-3-small",
    input=["Azure AI Foundry is a unified AI platform."]
)
vector = response.data[0].embedding

Azure AI Search

from azure.search.documents import SearchClient
from azure.search.documents.indexes import SearchIndexClient

# Create index
index_client = SearchIndexClient(endpoint, credential)
index_client.create_index(index_definition)

# Search with hybrid + semantic
search_client = SearchClient(endpoint, "docs-index", credential)
results = search_client.search(
    search_text="What is Foundry?",
    query_type="semantic",
    vector_queries=[{ "vector": query_vector, "fields": ["content_vector"] }],
    top=5
)

Multi-Modal RAG

Extend RAG to handle images, tables, and PDFs. Use Document Intelligence to extract content from documents before chunking and indexing.

Deploying Infrastructure

Use Bicep or Terraform to provision:

• Azure AI Search service
• Storage account for document upload
• AI Services for Document Intelligence
• Foundry Hub and Project