Vector Embeddings

Vector embeddings are numerical arrays (vectors) that represent the semantic meaning of text in high-dimensional space. Words or sentences with similar concepts cluster together geometrically, enabling AI systems to find relevant answers even when different vocabulary is used.

Detailed Explanation

Instead of matching exact keywords like traditional database search (e.g. 'return' vs 'refund'), embedding models convert text into lists of numbers (coordinates). CustomerGPT utilizes OpenAI's text-embedding-3-small model to map documents into 1536-dimensional vector space, allowing cosine similarity queries to identify semantic matches in under 50 milliseconds.

How It Works in CustomerGPT

When a shopper asks 'How do I send this jacket back?', CustomerGPT's vector embeddings immediately connect the query to your indexed 'Exchange & Return Policy' chunk without requiring keyword matching.

Related Glossary Terms

Retrieval-Augmented Generation (RAG) →Cosine Similarity →

Frequently Asked Questions

How many dimensions do CustomerGPT vector embeddings use?

CustomerGPT uses 1536-dimensional dense vectors generated by OpenAI text-embedding-3-small, indexed using pgvector with HNSW graphs.

Deploy Grounded RAG AI for Your Support

Get started in under 2 minutes with our 14-day free trial.

Start Free Trial →