04-02-03 — Azure AI Search como vector store

O que é Azure AI Search?

Azure AI Search (antigo "Azure Cognitive Search") é um serviço de search-as-a-service da Microsoft que evoluiu muito além de busca textual. Desde 2023 ganhou capacidades vetoriais de primeira classe e virou o vector store padrão do ecossistema Azure AI.

Capacidades relevantes para RAG:

• Vector search: HNSW com métrica cosine, dotProduct ou euclidean
• Full-text search: BM25 nativo, analisadores de linguagem para português
• Hybrid search: vetorial + BM25 com Reciprocal Rank Fusion automático
• Semantic ranker: reranking baseado em modelo de linguagem da Microsoft
• Integrated vectorization: chama Azure OpenAI Embeddings automaticamente na indexação
• Skillsets: pipeline de enriquecimento (OCR, extração de entidades, chunking) via Azure AI Services

Arquitetura com Azure AI Search

Criando um índice vetorial

Para configuração programática com o SDK Python:

from azure.search.documents.indexes import SearchIndexClient
from azure.search.documents.indexes.models import (
    SearchIndex, SearchField, SearchFieldDataType,
    VectorSearch, HnswAlgorithmConfiguration,
    VectorSearchProfile, SemanticConfiguration,
    SemanticSearch, SemanticPrioritizedFields, SemanticField
)
from azure.core.credentials import AzureKeyCredential

index_client = SearchIndexClient(
    endpoint="https://seu-servico.search.windows.net",
    credential=AzureKeyCredential("SUA_ADMIN_KEY")
)

fields = [
    SearchField(name="id", type=SearchFieldDataType.String, key=True),
    SearchField(name="content", type=SearchFieldDataType.String, searchable=True),
    SearchField(name="source", type=SearchFieldDataType.String, filterable=True),
    SearchField(
        name="content_vector",
        type=SearchFieldDataType.Collection(SearchFieldDataType.Single),
        searchable=True,
        vector_search_dimensions=1536,
        vector_search_profile_name="hnsw-profile"
    )
]

vector_search = VectorSearch(
    algorithms=[HnswAlgorithmConfiguration(name="hnsw-config", parameters={"m": 4, "efConstruction": 400})],
    profiles=[VectorSearchProfile(name="hnsw-profile", algorithm_configuration_name="hnsw-config")]
)

semantic_config = SemanticConfiguration(
    name="my-semantic-config",
    prioritized_fields=SemanticPrioritizedFields(content_fields=[SemanticField(field_name="content")])
)

index = SearchIndex(
    name="rag-index",
    fields=fields,
    vector_search=vector_search,
    semantic_search=SemanticSearch(configurations=[semantic_config])
)

index_client.create_or_update_index(index)
print("Índice criado!")

Indexando documentos

from azure.search.documents import SearchClient

search_client = SearchClient(
    endpoint="https://seu-servico.search.windows.net",
    index_name="rag-index",
    credential=AzureKeyCredential("SUA_ADMIN_KEY")
)

documents = []
for i, chunk in enumerate(chunks):
    embedding = embed(chunk)  # sua função de embedding
    documents.append({
        "id": f"chunk-{i}",
        "content": chunk,
        "source": "contrato.pdf",
        "content_vector": embedding
    })

search_client.upload_documents(documents)
print(f"{len(documents)} chunks indexados")

Buscando com hybrid + semantic reranker

from azure.search.documents.models import VectorizedQuery

def search_hybrid(question: str, k: int = 5) -> list[dict]:
    query_vector = embed(question)

    results = search_client.search(
        search_text=question,          # keyword search (BM25)
        vector_queries=[VectorizedQuery(
            vector=query_vector,
            k_nearest_neighbors=50,    # busca 50 candidatos vetoriais
            fields="content_vector"
        )],
        query_type="semantic",         # ativa semantic reranker
        semantic_configuration_name="my-semantic-config",
        top=k
    )

    return [{"content": r["content"], "source": r["source"], "score": r["@search.score"]}
            for r in results]

Tiers e custos