str.partition

rank=10
    ):
        start = time.time()
        query_embedding = self.model.encode([query], show_progress_bar=False)
        similarity = self.model.similarity_func(query_embedding, self.pooled_embeddings[lang])
        unsorted_index = np.argpartition(similarity, -rank)[-rank:]
        # unsorted_index = np.argsort(similarity)
        sorted_index = unsorted_index[np.flip(np.argsort(similarity[unsorted_index]))]
        results = []
        for index in sorted_index:
            results.append(
                {
                    'index': str(index),
                    **self.get_repo(index, lang),
                    'metric': 'similarity',
                    'value': float(similarity[index])
                }
            )
        end = time.time()
        runtime = float(end - start)
        return results, runtime    

class GitHubSearcher:
    # Comparison searcher using GitHub's API to perform plain text search
    def __init__(

max_repos: int = 100_000,
        use_rest_api: bool = False  # CHANGED: Default to GraphQL as per assignment
    ) -> dict:
        """
        Crawl GitHub repositories and save them to the database.
        Uses GraphQL API with partitioned queries to overcome 1K result limit.

        Args:
            queries: List of GitHub search queries (for GraphQL mode)
            max_repos: Maximum number of repositories to crawl across all queries
            use_rest_api: Use REST API (True) or GraphQL (False)

        Returns:
            Dictionary with crawl statistics
        """
        if queries is None:
            queries = self._get_optimized_queries_for_100k()

        print("\n" + "="*60)
        print(f"🚀 Starting GitHub Repository Crawler")
        print("="*60)
        print(f"📊 Target: {max_repos:,} repositories")
        print(f"🔍 API: {'REST' if use_rest_api else 'GraphQL (as required)'}")
        print(f"🔍 Query strategy: {len(queries)} partitioned queries")
        print(f"🗄️  Database: Ready")