Hello everyone,
I’m currently building an AI chatbot (with memory) for sales use case in Vietnamese that:
• Strictly follows given prompt (e.g. always polite, helpful, goal-driven to close a sale).
• Remembers conversation history to improve contextual responses
Data: dataset of real sales dialogues (between salespeople and customers). (cấu trúc như nào, định dạng gì, thì chèn cái ảnh vô)
My question:
• Full pipeline to deploy my app?
• Some rcm for each stage of the pipeline?
• My friend rcm to use RAG, is it possible?
Tks a lot for helping me <3
1 Like
For that purpose, I would also recommend RAG. If you require logical reasoning or human-like realism in responses, a large brain (large LLM) is necessary, but that’s probably not the case here. For response accuracy, the RAG approach, which retrieves information from a database, is more suitable as it is cost-effective and offers better database maintainability.
Additionally, with RAG, even the default settings of existing open-source LLMs with some intelligence can achieve decent performance. Depending on your requirements, some users may find a 3B model sufficient.
2 Likes