Fine-tuning LLM model for E-commerce Chatbot recomendation

I’d like to build a chatbot application to recommend E-commerce product given a list of previous actions taken (eg product X has been seen, product Y has been pushed in basket etc…).

What would be the best approach to achieve such objective ?
Last constraint :slight_smile: => I’d like this bot to answer in French

From my understanding so far, you can fine-tune a LLM model and map the “past behavior” of a user as a context for the model.