Skip to main content

​
Semantic Document Search

This example proves how easy it is to replace complex search infrastructure (like Elasticsearch or Algolia) with Jabrod’s native vector search capabilities.

​
Architecture

Traditional search engines match keywords. Semantic search understands intent.

​
SDK Implementation

The core of this example is the queryBuilder, which provides a fluent API for searching knowledge bases.

​
1. Ingesting Content

We upload purely text content. Jabrod automatically splits this into searchable segments. 
await client.kb.uploadText({
    kbName: 'docs-v1',
    content: documentText,
    metadata: { title: 'User Guide', version: '2.0' }
});

​
2. Searching

Instead of complex query DSLs, we just pass the natural language query. 
const results = await client.rag.queryBuilder()
    .withKB('docs-v1')
    .withQuery("how to fix payment errors")
    .withLimit(5) // Get top 5 matches
    .execute();

​
3. Displaying Results

The results contain the content snippet and the relevance score (0-1). 
results.forEach(match => {
    console.log(`Relevance: ${match.score}`);
    console.log(match.content);
});

​
Why it’s better

  1. No synonyms needed: “Billing” matches “Payment” automatically.
  2. No infrastructure: No vector DB to manage.
  3. Fast setup: Literal drag-and-drop of text.

View on GitHub

See the minimal search implementation in ex2-doc-search