Case 1. Semantic search over text documents (no DB)

Implementation in pure PHP

In this case we build a minimal semantic search in PHP: documents are transformed into embeddings, saved into a simple JSON index, and then ranked by cosine similarity to the query. This demonstrates the core engineering cycle (indexing -> query embedding -> similarity search -> top-N results) without a database or vector store.

Example of use


                
<?php

include 'code-en.php';

$query = 'How to add artificial intelligence to a PHP application?';
$result = $embedder($query, normalize: true, pooling: 'mean');
$queryEmbedding = array_map(static fn ($v): float => (float) $v, $result[0]);

$scored = [];

foreach ($documents as $document) {
    $scored[] = [
        'score' => cosineSimilarity($queryEmbedding, $document['embedding']),
        'document' => $document,
    ];
}

usort($scored, static fn (array $a, array $b): int => $b['score'] <=> $a['score']);
$topResults = array_slice($scored, 0, 3);

echo 'Query: ' . $query . PHP_EOL . PHP_EOL;

if (count($topResults) === 0) {
    echo 'No results found.' . PHP_EOL;
    return;
}

foreach ($topResults as $row) {
    $score = number_format((float) $row['score'], 2, '.', '');
    $document = $row['document'];

    echo '[' . $score . '] ' . $document['id'] . PHP_EOL;
    echo (string) $document['text'] . PHP_EOL . PHP_EOL;
}

Result: Memory: 0.001 Mb Time running: < 0.001 sec.


                Query: How to add artificial intelligence to a PHP application?

[0.71] php-ai.md
PHP is gradually becoming part of AI infrastructure.
Developers use PHP to build applications
with language models and intelligent features.

[0.36] machine-learning.txt
Machine learning helps build models
that detect patterns in data
and improve search and recommendation quality.

[0.33] laravel.md
Laravel is used to build
modern web applications.