卡在一個獨特的自然語言處理問題上 [D]

So basically, I am developing an app where I would need to classify the texts. The problem is the texts can be in English, Hindi and hindi+english(Hindi language written with English alphabets). So naturally I chose the way of sentence transformer for it but the main problem is it fails abysmally on Hindi+English. There seems to be zero semantic meaning to the model of these type of tasks. I know LLM is a solution for this but my application would be too heavy with it. I thought of transliterati