name_matching/inference.py

import os
from typing import Optional, List

import spacy
from huggingface_hub import snapshot_download

hf_token: str = os.getenv('HF_TOKEN')


class NameMatchingInference:
    @classmethod
    def get_name_matches(cls, payer: str, company_id: str):
        model_repo: str = f"siva-valyx/name-matching-model-{company_id}"
        # Download the model using Hugging Face's API
        model_path: str = snapshot_download(repo_id=model_repo, revision="main", token=hf_token)

        # Load the model in spaCy
        nlp = spacy.load(model_path)

        doc = nlp(payer)

        sorted_cats = sorted(doc.cats.items(), key=lambda item: item[1], reverse=True)

        top_matches: List[str] = []

        top_score: float = sorted_cats[0][1]

        if top_score < 0.03:
            return top_matches

        similarity_threshold: float = top_score * 0.3

        for customer_id, score in sorted_cats:

            if top_score - score > similarity_threshold:
                break

            top_matches.append(customer_id)

        return top_matches