const logger = require('../utils/logger');

let pipeline;
try {
  // Lazy import to avoid startup cost when unused
  ({ pipeline } = require('@xenova/transformers'));
} catch (e) {
  logger.warn('Embedding pipeline not available. Did you install @xenova/transformers?');
}

class EmbeddingService {
  constructor() {
    this.initialized = false;
    this.extractor = null;
    this.modelName = process.env.EMBEDDING_MODEL || 'Xenova/all-MiniLM-L6-v2';
  }

  async initIfNeeded() {
    if (this.initialized) return;
    if (!pipeline) {
      throw new Error('Transformers pipeline not available');
    }
    this.extractor = await pipeline('feature-extraction', this.modelName);
    this.initialized = true;
    logger.info(`Embedding model loaded: ${this.modelName}`);
  }

  async embedText(text) {
    if (!text || !text.trim()) return [];
    await this.initIfNeeded();
    const output = await this.extractor(text, { pooling: 'mean', normalize: true });
    // output is a Tensor; convert to plain JS array
    // Depending on version, .data or .tolist()
    const vector = Array.isArray(output) ? output : (output?.data ? Array.from(output.data) : output.tolist());
    return vector;
  }

  cosineSimilarity(a, b) {
    if (!a || !b || a.length !== b.length || a.length === 0) return 0;
    let dot = 0;
    let normA = 0;
    let normB = 0;
    for (let i = 0; i < a.length; i++) {
      const va = a[i] || 0;
      const vb = b[i] || 0;
      dot += va * vb;
      normA += va * va;
      normB += vb * vb;
    }
    const denom = Math.sqrt(normA) * Math.sqrt(normB);
    return denom ? dot / denom : 0;
  }
}

module.exports = new EmbeddingService();