AdryanAzul's picture
feat: deploy DocuMind AI (RAG + BYOK + local embeddings)
68e8c03 verified
Raw History Blame Contribute Delete
1.13 kB
"""Recuperação semântica de trechos relevantes no banco vetorial."""
from src.embeddings import EmbeddingModel
from src.vector_store import StoredDocument, VectorStore
class Retriever:
"""Realiza busca semântica combinando o modelo de embeddings e o VectorStore."""
def __init__(self, embedding_model: EmbeddingModel, vector_store: VectorStore) -> None:
self.embedding_model = embedding_model
self.vector_store = vector_store
def retrieve(
self,
query: str,
top_k: int = 4,
max_distance: float | None = 0.85,
) -> list[StoredDocument]:
"""Recupera os top_k trechos mais relevantes para a pergunta informada."""
cleaned_query = query.strip()
if not cleaned_query:
return []
query_embedding = self.embedding_model.embed_query(cleaned_query)
docs = self.vector_store.query(query_embedding=query_embedding, top_k=top_k)
if max_distance is None:
return docs
return [
doc for doc in docs
if doc.distance is None or doc.distance <= max_distance
]