How to Intermediate · 4 min read

How to implement semantic product search

Quick answer

Implement semantic product search by encoding product descriptions and user queries into vector embeddings using models like text-embedding-3-small. Then perform similarity search with vector databases such as FAISS to retrieve relevant products based on semantic meaning rather than keywords.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0 faiss-cpu numpy

Setup

Install required Python packages and set your OpenAI API key as an environment variable.

bash

pip install openai faiss-cpu numpy

# Set environment variable in your shell
export OPENAI_API_KEY=os.environ["OPENAI_API_KEY"]

output

Requirement already satisfied: openai in ...
Requirement already satisfied: faiss-cpu in ...
Requirement already satisfied: numpy in ...

Step by step

This example encodes product descriptions and a user query into embeddings, then uses FAISS to find the top 3 semantically similar products.

python

import os
import numpy as np
import faiss
from openai import OpenAI

# Initialize OpenAI client
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Sample product catalog
products = [
    {"id": 1, "name": "Wireless Bluetooth Headphones", "description": "High quality wireless headphones with noise cancellation."},
    {"id": 2, "name": "Running Shoes", "description": "Lightweight running shoes with breathable mesh."},
    {"id": 3, "name": "Smartphone Case", "description": "Durable case compatible with latest smartphone models."},
    {"id": 4, "name": "Electric Toothbrush", "description": "Rechargeable toothbrush with multiple cleaning modes."}
]

# Function to get embeddings from OpenAI
def get_embedding(text):
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=text
    )
    return np.array(response.data[0].embedding, dtype=np.float32)

# Create embeddings for all products
product_embeddings = np.array([get_embedding(p["description"]) for p in products])

# Build FAISS index
dimension = product_embeddings.shape[1]
index = faiss.IndexFlatL2(dimension)
index.add(product_embeddings)

# User query
query = "Looking for noise cancelling headphones"
query_embedding = get_embedding(query)

# Search top 3 similar products
k = 3
D, I = index.search(np.array([query_embedding]), k)

print("Top products for query:", query)
for rank, idx in enumerate(I[0]):
    product = products[idx]
    print(f"{rank + 1}. {product['name']} - {product['description']}")

output

Top products for query: Looking for noise cancelling headphones
1. Wireless Bluetooth Headphones - High quality wireless headphones with noise cancellation.
2. Electric Toothbrush - Rechargeable toothbrush with multiple cleaning modes.
3. Smartphone Case - Durable case compatible with latest smartphone models.

Common variations

Use async calls with asyncio and OpenAI's async client for higher throughput.
Switch to other embedding models like claude-3-5-sonnet embeddings or sentence-transformers for local inference.
Use vector databases like Pinecone or Chroma for scalable, persistent search.

Troubleshooting

If embeddings are slow, batch multiple texts per request to OpenAI.
If search results are irrelevant, verify embedding model choice and consider fine-tuning or prompt engineering.
Ensure faiss index dimension matches embedding size exactly to avoid errors.

✅

Key Takeaways

Use vector embeddings to represent product descriptions and queries for semantic similarity.
FAISS provides fast in-memory similarity search for small to medium catalogs.
OpenAI's text-embedding-3-small is a reliable embedding model for semantic search.
For production, consider scalable vector DBs like Pinecone or Chroma.
Batch embedding requests and verify index dimensions to optimize performance.

Verified 2026-04 · text-embedding-3-small, gpt-4o, claude-3-5-sonnet-20241022

Verify ↗