/**
 * TF-IDF Embedding Provider — zero-dependency semantic search
 * (Term Frequency–Inverse Document Frequency)
 * Provides a lightweight text-to-vector implementation using TF-IDF (Term
 * Frequency–Inverse Document Frequency) for cosine-similarity ranking. This
 * is good enough for 10–200 skills. For production enterprise deployments,
 * plug in an OpenAI/Cohere `EmbeddingProvider` instead.
 *
 * No external dependencies — works out of the box.
 */
import { EmbeddingProvider } from './types.js';
/**
 * Simple TF-IDF based embedding provider.
 * Builds a vocabulary from the registered corpus and represents each text as
 * a TF-IDF weighted vector.
 */
export declare class TfIdfEmbeddingProvider implements EmbeddingProvider {
    private vocabulary;
    private idf;
    private corpusSize;
    private _dimensions;
    get dimensions(): number;
    /**
     * Build the vocabulary and IDF weights from a corpus of documents.
     * Must be called before `embed()` or `embedBatch()`.
     */
    fit(documents: string[]): void;
    embed(text: string): Promise<number[]>;
    embedBatch(texts: string[]): Promise<number[][]>;
    /**
     * Synchronous embed for internal use (no async overhead).
     */
    embedSync(text: string): number[];
    /**
     * Tokenize text into lowercase terms.
     * Splits on non-alphanumeric characters and filters stopwords.
     */
    private tokenize;
}
/**
 * Cosine similarity between two vectors.
 * Returns a value between -1 and 1 (1 = identical, 0 = orthogonal).
 */
export declare function cosineSimilarity(a: number[], b: number[]): number;
//# sourceMappingURL=tfidf-embedding.d.ts.map