Real World Example

Flipkart, Myntra and Meesho let shoppers search in their own words ("shaadi ke liye red lehenga under 5000"). Behind it: product text is embedded, stored with price/stock/category payloads, and searched with filters. A ShopEase version with Qdrant:

from qdrant_client import QdrantClient, models
import voyageai

vo = voyageai.Client()
qdrant = QdrantClient(url="http://localhost:6333")

qdrant.create_collection(
    collection_name="products",
    vectors_config=models.VectorParams(size=1024, distance=models.Distance.COSINE),
)

products = load_products()   # [{"id": 1, "title": ..., "description": ..., "price": ..., "in_stock": ...}]
vectors = vo.embed([p["title"] + ". " + p["description"] for p in products],
                   model="voyage-3.5", input_type="document").embeddings
qdrant.upsert(collection_name="products", points=[
    models.PointStruct(id=p["id"], vector=v, payload={"title": p["title"], "price": p["price"], "in_stock": p["in_stock"]})
    for p, v in zip(products, vectors)
])

def search(query: str, max_price: int):
    qv = vo.embed([query], model="voyage-3.5", input_type="query").embeddings[0]
    hits = qdrant.query_points(
        collection_name="products", query=qv, limit=3,
        query_filter=models.Filter(must=[
            models.FieldCondition(key="price", range=models.Range(lte=max_price)),
            models.FieldCondition(key="in_stock", match=models.MatchValue(value=True)),
        ]),
    ).points
    return [(round(h.score, 3), h.payload["title"], h.payload["price"]) for h in hits]

print(search("red lehenga for a wedding", max_price=5000))

python product_search.py

[(0.812, 'Maroon silk lehenga choli with dupatta', 4599),
 (0.794, 'Red bridal georgette lehenga, embroidered', 4999),
 (0.731, 'Rani pink festive lehenga set', 3899)]

• The price limit comes from parsing the query ("under 5000"), often with a small LLM call using structured output.
• Shops mix semantic scores with keyword search, popularity and personalization (hybrid search is covered in the advance level).
• Re-embed a product when its title or description changes.