Class HybridUtilityAI

Routes each utility method to the most appropriate backend:

  • LLM: summarization, classification, keyword extraction, JSON repair
  • Statistical: tokenization, stemming, n-grams, readability, similarity
  • Either with preference: sentiment, language detection

A summary, classification, keyword list, similarity or sentiment whose method names a method only one class has goes to that class when the hybrid holds it. The statistical class takes first_n_sentences and extractive_sentence_rank summaries, naive_bayes and keyword_matching classification, tf_idf and frequency_based keywords, cosine_tfidf, jaccard and levenshtein similarity and lexicon_based sentiment; the LLM-backed class takes abstractive_llm and key_points_llm summaries, llm_zeroshot classification, llm keywords, llm_semantic similarity and llm sentiment. If the preferred backend is not held, the other one takes the call.

Implements

Constructors

constructor

Methods

initialize summarize classifyText extractKeywords parseJsonSafe tokenize stemTokens normalizeText generateNGrams calculateReadability calculateSimilarity analyzeSentiment detectLanguage checkHealth shutdown

Properties

utilityId

Constructors

constructor

Methods

initialize

summarize

classifyText

extractKeywords

parseJsonSafe

  • parseJsonSafe<T>(jsonString, options?): Promise<null | T>

  • Parses a string that is expected to be JSON: as it is, else from inside the Markdown code fence it is or holds, which a model asked for JSON alone often answers in. An LLM-backed class sends a text that still does not parse for repair when attemptFixWithLLM is set.

    Type Parameters

    • T = any

      The expected type of the parsed JSON object.

    Parameters

    • jsonString: string

      The string to parse.

    • Optional options: ParseJsonOptions<T>

      Options for parsing and fixing.

    Returns Promise<null | T>

    The parsed object, or null if parsing and fixing fail.

tokenize

stemTokens

normalizeText

generateNGrams

calculateReadability

calculateSimilarity

analyzeSentiment

detectLanguage

checkHealth

  • checkHealth(): Promise<{
        isHealthy: boolean;
        details?: any;
        dependencies?: {
            name: string;
            isHealthy: boolean;
            details?: any;
        }[];
    }>

  • Returns Promise<{
        isHealthy: boolean;
        details?: any;
        dependencies?: {
            name: string;
            isHealthy: boolean;
            details?: any;
        }[];
    }>

shutdown

Properties

Readonly utilityId

utilityId: string

Settings

Member Visibility

Theme

On This Page

constructorinitializesummarizeclassifyTextextractKeywordsparseJsonSafetokenizestemTokensnormalizeTextgenerateNGramscalculateReadabilitycalculateSimilarityanalyzeSentimentdetectLanguagecheckHealthshutdownutilityId