Understanding Semantic Search and Embeddings

Semantic Search

Helpium uses AI-powered semantic search to help customers find relevant articles, even if they don't use the exact words in the article title.

How It Works

When you publish an article, Helpium automatically:

1. Splits the article content into smaller chunks (approximately 500 tokens each)
2. Generates a vector embedding for each chunk using OpenAI's embedding model
3. Stores the embeddings in a pgvector-enabled PostgreSQL database

When a customer searches, their query is also converted into an embedding. Helpium then finds the most similar article chunks using vector similarity search.

Why Semantic Search Matters

Traditional keyword search only matches exact words. Semantic search understands meaning:

• Searching "how to change my password" finds articles about "updating credentials" or "resetting your login"
• Searching "billing problem" finds articles about "invoice issues" or "payment errors"

Automatic Embedding Updates

Embeddings are regenerated automatically when you:

• Publish an article
• Update a published article's content, title, or excerpt
• Change an article's status to published

Fallback to Keyword Search

If no embeddings are available (e.g., for a new workspace with no published articles), Helpium falls back to keyword-based search using PostgreSQL's case-insensitive matching.

Search Suggestions

The help center and widget provide typeahead search suggestions as customers type. These are generated using prefix matching on article titles for fast, real-time results.