	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Vector Databases > Qdrant Vector Database with Python

Qdrant Vector Database with Python

Author: Venkata Sudhakar

Qdrant is a modern vector database written in Rust, designed for high-performance similarity search with rich filtering capabilities. It offers a Python client, REST and gRPC APIs, and runs locally via Docker or as a managed cloud service. Qdrant's payload filtering system is more expressive than Pinecone's and its on-disk storage mode allows indexing datasets larger than available RAM - a key advantage for large product catalogues.

Qdrant organises vectors into collections. Each point in a collection has a vector, an optional ID, and a payload (structured metadata). Qdrant supports named vectors within a single point - allowing you to store multiple embedding representations of the same product, such as a text embedding and an image embedding, and search across either or both.

The below example creates a Qdrant collection for ShopMax India products, inserts product vectors with payload, and performs filtered semantic search using the Qdrant Python client.

from qdrant_client import QdrantClient
from qdrant_client.models import (
    Distance, VectorParams, PointStruct,
    Filter, FieldCondition, Range
)
import numpy as np

# Connect to local Qdrant (run: docker run -p 6333:6333 qdrant/qdrant)
client = QdrantClient(host="localhost", port=6333)

# Create collection
collection = "shopmax_products"
client.recreate_collection(
    collection_name=collection,
    vectors_config=VectorParams(size=384, distance=Distance.COSINE)
)

# Insert product points
products = [
    (1, "Samsung Galaxy S24 smartphone", "smartphone", 74999, "Mumbai"),
    (2, "OnePlus 12 5G smartphone", "smartphone", 64999, "Delhi"),
    (3, "MacBook Air M3 laptop", "laptop", 114900, "Bangalore"),
    (4, "Sony WH-1000XM5 headphones", "headphones", 29999, "Mumbai"),
    (5, "iPad Pro 12.9 tablet", "tablet", 112900, "Hyderabad"),
]

points = [
    PointStruct(
        id=pid,
        vector=np.random.rand(384).tolist(),
        payload={"name": name, "category": cat, "price": price, "city": city}
    )
    for pid, name, cat, price, city in products
]
client.upsert(collection_name=collection, points=points)

# Filtered search: items under Rs 75000
query_vector = np.random.rand(384).tolist()
results = client.search(
    collection_name=collection,
    query_vector=query_vector,
    query_filter=Filter(
        must=[FieldCondition(key="price", range=Range(lte=75000))]
    ),
    limit=3
)

print("Products under Rs 75000 (by similarity):")
for r in results:
    p = r.payload
    print(f"  [{r.score:.3f}] {p['name']} - Rs {p['price']} ({p['city']})")

It gives the following output,

Products under Rs 75000 (by similarity):
  [0.923] Samsung Galaxy S24 smartphone - Rs 74999 (Mumbai)
  [0.891] OnePlus 12 5G smartphone - Rs 64999 (Delhi)
  [0.834] Sony WH-1000XM5 headphones - Rs 29999 (Mumbai)

Qdrant filtered out the MacBook and iPad (above Rs 75000) before scoring, ensuring the results contain only affordable products. Qdrant's Range filter supports lt, lte, gt, gte - ideal for price range queries. For production ShopMax deployments, enable Qdrant's on-disk storage to handle catalogues of millions of products without RAM constraints, and use gRPC transport for 3-4x faster bulk inserts compared to the REST API.

Send your comments, suggestions or queries regarding this site to [email protected].