backend/agents/matcher.py

from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
import streamlit as st

# Use a singleton pattern to load the model only once.
@st.cache_resource
def get_sentence_transformer_model():
    """Loads and caches the SentenceTransformer model."""
    try:
        model = SentenceTransformer('all-MiniLM-L6-v2')
        return model
    except Exception as e:
        st.error(f"Error loading embedding model: {e}")
        return None

def get_embedding(text: str, api_key: str = None, model_name: str = None):
    """
    Generates embeddings for a given text using a local SentenceTransformer model.
    The api_key and model_name arguments are kept for compatibility but are no longer used.

    Args:
        text (str): The input text to embed.

    Returns:
        np.ndarray: A numpy array representing the embedding.
    """
    model = get_sentence_transformer_model()
    if model is None:
        return None
        
    try:
        # The model directly outputs a numpy array.
        embedding = model.encode(text)
        return embedding
    except Exception as e:
        print(f"Error generating embedding: {e}")
        raise

def calculate_match_score(resume_embedding: np.ndarray, job_embedding: np.ndarray) -> float:
    """
    Calculates the cosine similarity between two embeddings.

    Args:
        resume_embedding (np.ndarray): The embedding of the resume.
        job_embedding (np.ndarray): The embedding of the job description.

    Returns:
        float: The cosine similarity score (0.0 to 1.0).
    """
    if resume_embedding is None or job_embedding is None:
        return 0.0

    # Reshape arrays to be 2D for cosine_similarity function
    resume_embedding = resume_embedding.reshape(1, -1)
    job_embedding = job_embedding.reshape(1, -1)

    # Calculate cosine similarity
    score = cosine_similarity(resume_embedding, job_embedding)[0][0]

    # Ensure the score is within the 0-1 range
    return max(0.0, min(1.0, score))