text-embedding-3-small

Real World Example

Illustrative business scenario · AI-generated from linked documentation · Not a measured customer case study

Optimizing Internal Knowledge Retrieval for Global Logistics

A global logistics firm managing thousands of unstructured shipping manifests, regulatory documents, and internal process manuals.

The business problem

Employees struggle to locate specific compliance procedures within a massive, growing repository of text-heavy documents, leading to operational delays and inconsistent adherence to regional shipping regulations.

How this topic solves it

The firm implements text-embedding-3-small to convert diverse internal documents into high-dimensional vector representations. By indexing these embeddings in a vector database, the system enables semantic search, allowing employees to query for concepts like 'hazardous material handling in EU ports' rather than relying on exact keyword matches.

1. Pre-process and chunk internal documents into manageable text segments.
2. Send text chunks to the v1/embeddings endpoint using the text-embedding-3-small model.
3. Store the resulting numerical vector arrays in a vector-capable database.
4. Convert user search queries into embeddings using the same model.
5. Perform a cosine similarity search to retrieve the most relevant document chunks for the user.

Implementation example

This code demonstrates calling the v1/embeddings endpoint to transform a text string into a vector, which is the core requirement for the semantic search solution.

python · Illustrative code; not executed or production-validated.

import openai
# Generate embeddings for a document chunk
response = openai.Embedding.create(
    input="EU port hazardous material protocol",
    model="text-embedding-3-small" # Specify the model ID
)
vector = response['data'][0]['embedding'] # Extract the numerical representation

Expected business outcome

Employees experience a significant reduction in time spent searching for information, as the system surfaces contextually relevant documents even when the user's terminology differs from the source text.

Limitations and checks

The model produces fixed-length vector representations. A primary limitation is that while these embeddings capture semantic meaning, they do not inherently understand complex logical reasoning or multi-step document synthesis. To verify if the model is sufficient for a specific use case, developers should perform an evaluation using a golden dataset of query-document pairs to measure retrieval precision.

Sources

• text-embedding-3-small

Generated 2026-10-18 with gemini-3.1-flash-lite. Your assessment includes two questions about this example.