embeddings/model_manager.py

from sentence_transformers import SentenceTransformer
from typing import List
import numpy as np

class EmbeddingModel:
    """Lightweight sentence-transformer for semantic similarity"""
    
    def __init__(self, model_name: str = "paraphrase-MiniLM-L3-v2"):
        # Use smallest model (3 layers, ~60MB, ~100MB RAM)
        self.model = SentenceTransformer(model_name)
        self.model_name = model_name
        self.dimension = self.model.get_sentence_embedding_dimension()
    
    def encode(self, texts: List[str], batch_size: int = 32) -> np.ndarray:
        """Generate embeddings for a list of texts"""
        if isinstance(texts, str):
            texts = [texts]
        
        cleaned_texts = []
        for text in texts:
            if text and isinstance(text, str):
                cleaned = ' '.join(text.split())
                if cleaned.strip():
                    cleaned_texts.append(cleaned)
        
        if not cleaned_texts:
            return np.array([])
        
        embeddings = self.model.encode(cleaned_texts, batch_size=batch_size, convert_to_numpy=True)
        return embeddings
    
    def encode_single(self, text: str) -> np.ndarray:
        """Generate embedding for a single text"""
        cleaned = ' '.join(text.split()) if text else ""
        if not cleaned.strip():
            return np.array([])
        
        embedding = self.model.encode([cleaned], convert_to_numpy=True)
        return embedding[0]