ARTICLE AD BOX
I am building a stock assistant chatbot using an LLM.
I want the chatbot to always respond in Hinglish, i.e. a mix of Hindi and English written in Roman script, for example: Aaj market thoda volatile hai, lekin long-term ke liye ye stock achha lag raha hai.
Currently, the model replies only in English.
What I want to achieve
Responses should be Hinglish by default
Hindi words should be in Roman script, not Devanagari
The chatbot should keep the financial terms in English where appropriate
What I’ve tried
Setting system prompts like:
You are a stock assistant who answers in Hinglish
Passing user instructions asking to respond in Hinglish
However, the output is still inconsistent and sometimes fully English.
My questions
Is prompt engineering alone enough to enforce Hinglish responses?
Is there a recommended way (system prompt, few-shot examples, output parser, etc.) to make the model consistently reply in Hinglish?
If using LangChain, is there any best practice or pattern for multilingual / mixed-language responses?
