I’m a beginner in AI, so please correct me if I’m wrong or missing something obvious. I’m trying to learn and would really appreciate your help.
I’m building a chatbot for my SaaS low-code/no-code platform where users can design applications using drag-and-drop tools and custom configurations. Currently, I use a Retrieval-Augmented Generation (RAG) approach to let the bot answer “how-to” and “what-is” style questions, which works for general documentation and feature explanations.
However, the core challenge is this:
My users are developing applications inside the platform—for example, creating a Hospital Patient Management app. These use cases require domain-specific logic, like which fields to include, what workflows to design, what triggers to set, etc. These are not static answers but involve reasoning based on both platform capabilities and the app’s domain.
I’ve considered fine-tuning, but that adjusts existing model weights rather than adding truly new domain knowledge or logic. So fine-tuning alone doesn’t solve the problem.
What I really need is a solution where the chatbot can help users design apps contextually based on:
-
What kind of app they want to create (e.g., patient management,
inventory, CRM) -
The available tools in the platform (Forms, Workflows, Datasets,
Reports, etc.) -
Logical reasoning to generate recommendations, field structures, and
flows
What I’ve tried:
-
RAG with embedded documentation and examples
-
Fine-tuning with custom Q&A based on features
But still facing issues:
-
Lack of reasoning or “logical build” ability from the bot
-
No way to generalize across custom app types or domains
-
Chatbot can’t make recommendations like “Add these fields for patient
management,” “Use this workflow for appointment scheduling,” etc.
Any help, architecture suggestions, or examples would be appreciated.