Qdrant: Advanced Vector Search Features

Why Qdrant?

• Open-source & self-hosted
• Advanced filtering
• Scalar quantization (4x smaller)
• Distributed clustering
• Built-in sparse vectors

Docker Setup

DEVELOPERbash
docker run -p 6333:6333 qdrant/qdrant

DEVELOPERpython
from qdrant_client import QdrantClient
from qdrant_client.models import Distance, VectorParams

client = QdrantClient("localhost", port=6333)

# Create collection
client.create_collection(
    collection_name="documents",
    vectors_config=VectorParams(
        size=1536,
        distance=Distance.COSINE
    )
)

Payload Indexing

Index metadata for fast filtering:

DEVELOPERpython
# Create index on "category" field
client.create_payload_index(
    collection_name="documents",
    field_name="category",
    field_schema="keyword"
)

# Create index on numeric "price"
client.create_payload_index(
    collection_name="documents",
    field_name="price",
    field_schema="integer"
)

# Now filtering is fast
results = client.search(
    collection_name="documents",
    query_vector=embedding,
    query_filter=Filter(
        must=[
            FieldCondition(key="category", match=MatchValue(value="tech")),
            FieldCondition(key="price", range=Range(lt=100))
        ]
    ),
    limit=10
)

Quantization (4x Compression)

DEVELOPERpython
from qdrant_client.models import ScalarQuantization, ScalarType, QuantizationSearchParams

# Enable quantization
client.update_collection(
    collection_name="documents",
    quantization_config=ScalarQuantization(
        type=ScalarType.INT8,
        quantile=0.99,
        always_ram=True
    )
)

# Search with quantization
results = client.search(
    collection_name="documents",
    query_vector=embedding,
    search_params=QuantizationSearchParams(
        ignore=False,  # Use quantized vectors
        rescore=True    # Rescore with full precision
    ),
    limit=10
)

Result: 1GB index → 256MB (4x smaller, 10% accuracy loss)

Distributed Deployment

DEVELOPERyaml
# docker-compose.yml
version: '3.8'
services:
  qdrant-node1:
    image: qdrant/qdrant
    environment:
      - QDRANT__CLUSTER__ENABLED=true
      - QDRANT__CLUSTER__P2P__PORT=6335
    ports:
      - "6333:6333"
    
  qdrant-node2:
    image: qdrant/qdrant
    environment:
      - QDRANT__CLUSTER__ENABLED=true
      - QDRANT__CLUSTER__P2P__PORT=6335
      - QDRANT__CLUSTER__BOOTSTRAP__P2P__URI=qdrant-node1:6335

Sparse Vectors (Hybrid Search)

DEVELOPERpython
from qdrant_client.models import SparseVector, NamedVector

# Upsert with both dense and sparse
client.upsert(
    collection_name="hybrid",
    points=[{
        "id": 1,
        "vector": {
            "dense": [0.1, 0.2, ...],  # Dense embedding
            "sparse": SparseVector(
                indices=[10, 45, 123],
                values=[0.5, 0.3, 0.2]
            )
        },
        "payload": {"text": "..."}
    }]
)

# Hybrid search
results = client.query_points(
    collection_name="hybrid",
    prefetch=[
        Prefetch(using="dense", query=[0.1, 0.2, ...], limit=100),
        Prefetch(using="sparse", query=SparseVector(...), limit=100)
    ],
    query=FusionQuery(fusion=Fusion.RRF),
    limit=10
)

Qdrant combines power, flexibility, and performance. Perfect for advanced RAG use cases.