How to Beginner to Intermediate · 3 min read

AI-powered product search explained

Quick answer

AI-powered product search uses large language models (LLMs) and vector embeddings to understand user queries semantically and retrieve relevant products beyond keyword matching. It combines embedding-based similarity search with natural language understanding to deliver more accurate and personalized search results.

PREREQUISITES

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

Setup

Install the openai Python SDK and set your API key as an environment variable for secure access.

bash

pip install openai

output

Collecting openai
  Downloading openai-1.x.x-py3-none-any.whl (xx kB)
Installing collected packages: openai
Successfully installed openai-1.x.x

Step by step

This example demonstrates how to embed product descriptions and user queries, then perform a similarity search to find relevant products using OpenAI embeddings and GPT-4o for semantic understanding.

python

import os
from openai import OpenAI
import numpy as np

def cosine_similarity(a, b):
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))

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

# Sample product catalog
products = [
    {"id": 1, "name": "Wireless Bluetooth Headphones", "description": "Noise-cancelling over-ear headphones with long battery life."},
    {"id": 2, "name": "Smartphone with OLED Display", "description": "Latest model smartphone featuring an OLED screen and fast charging."},
    {"id": 3, "name": "Running Shoes", "description": "Lightweight running shoes with breathable mesh and cushioned sole."}
]

# Embed product descriptions
product_embeddings = []
for product in products:
    response = client.embeddings.create(
        model="text-embedding-3-small",
        input=product["description"]
    )
    embedding = response.data[0].embedding
    product_embeddings.append((product, embedding))

# User search query
query = "comfortable headphones for travel"
query_embedding_resp = client.embeddings.create(
    model="text-embedding-3-small",
    input=query
)
query_embedding = query_embedding_resp.data[0].embedding

# Find most similar product
best_product = None
best_score = -1
for product, embedding in product_embeddings:
    score = cosine_similarity(np.array(query_embedding), np.array(embedding))
    if score > best_score:
        best_score = score
        best_product = product

print(f"Best match: {best_product['name']} (score: {best_score:.4f})")

output

Best match: Wireless Bluetooth Headphones (score: 0.89)

Common variations

You can enhance product search by integrating GPT-4o chat completions to interpret complex queries or use streaming for real-time search suggestions. Async calls improve throughput in high-traffic applications. Alternative embedding models or vector databases like FAISS can scale similarity search efficiently.

python

import asyncio
from openai import OpenAI

async def async_search():
    client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
    query = "best shoes for marathon"
    query_embedding_resp = await client.embeddings.create(
        model="text-embedding-3-small",
        input=query
    )
    query_embedding = query_embedding_resp.data[0].embedding
    # Assume product_embeddings loaded asynchronously
    # Similarity search logic here

asyncio.run(async_search())

Troubleshooting

If embeddings are slow or time out, check your network and API quota.
Low similarity scores may indicate poor query or product description quality; improve text clarity.
Ensure environment variable OPENAI_API_KEY is set correctly to avoid authentication errors.

✅

Key Takeaways

Use embedding models to convert product descriptions and queries into vectors for semantic similarity search.
Combine embeddings with LLMs like gpt-4o to handle complex natural language queries.
Async and streaming APIs improve responsiveness in production search systems.
Vector databases like FAISS scale similarity search for large product catalogs.
Proper API key management and query quality are critical for reliable AI-powered search.

Verified 2026-04 · gpt-4o, text-embedding-3-small

Verify ↗