File size: 3,939 Bytes
481dfe6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 | import json
import logging
from langchain_core.documents import Document
from app.retrieval.qdrant_store import get_qdrant_client, get_embeddings
from app.retrieval.reranker import rerank_documents
from app.config import settings
logger = logging.getLogger(__name__)
def retrieve_and_rerank(query: str) -> list[Document]:
"""
Embed query → search Qdrant with dense vectors (top_k) →
deserialize payloads → Cohere rerank (top_n).
Returns reranked documents with content and metadata.
"""
client = get_qdrant_client()
embeddings = get_embeddings()
# Get dense embedding for the query
try:
dense_embedding = embeddings.embed_query(query)
except Exception as e:
logger.error("Failed to embed query: %s", e)
return []
# Try to search in Qdrant
search_result = None
# Query using the official Qdrant API with 'using' parameter for named vectors
try:
search_result = client.query_points(
collection_name=settings.collection_name,
query=dense_embedding,
using=settings.vector_name, # Official Qdrant syntax for named vectors
limit=settings.retrieval_top_k,
with_payload=True,
with_vectors=False,
)
search_result = (
search_result.points if hasattr(search_result, 'points') else search_result
)
logger.debug("Query executed successfully with named vector: %s", settings.vector_name)
except Exception as e:
logger.error("Failed to query Qdrant: %s", e)
return []
if not search_result:
logger.warning("No search results returned from Qdrant")
return []
# Convert Qdrant results to LangChain Documents
documents = []
for scored_point in search_result:
payload = scored_point.payload or {}
# Deserialize _node_content if present
page_content = ""
metadata = {}
if "_node_content" in payload:
try:
node_data = json.loads(payload["_node_content"])
page_content = node_data.get("text", "")
# Preserve original metadata fields
if "metadata" in node_data:
metadata = node_data["metadata"]
except (json.JSONDecodeError, KeyError) as e:
logger.debug("Failed to deserialize node_content: %s", e)
# Add other payload fields as metadata
for key, value in payload.items():
if key not in ["_node_content"] and key not in metadata:
metadata[key] = value
doc = Document(page_content=page_content, metadata=metadata)
documents.append(doc)
if not documents:
logger.warning("No documents with content after deserialization")
return []
logger.info("Retrieved %d documents from Qdrant", len(documents))
# Rerank documents using Cohere
try:
reranked = rerank_documents(query=query, documents=documents)
logger.info("Reranked to %d documents", len(reranked))
return reranked
except Exception as e:
logger.error("Reranking failed: %s, returning original documents", e)
return documents
def format_sources(documents: list[Document]) -> list[dict]:
"""Extract source metadata from documents for the API response."""
sources = []
for i, doc in enumerate(documents):
meta = doc.metadata or {}
sources.append(
{
"rank": i + 1,
"score": float(meta.get("relevance_score", meta.get("score", 0.0))),
"file_name": meta.get("file_name", ""),
"url": meta.get("url", meta.get("page_id", "")),
"text_preview": doc.page_content[:300],
}
)
return sources
|