Building a chatbot with decision tree and specific question and answers

[PaulHuatingSun]

I am very new to both OpenAI and LangChain and I was recently tasked with creating a chatbot that use LLMs to answer user's questions. The expectation of the chatbot is that user will type in specific text and the bot will answer specific text based on the user input. Somewhat like a bot decision tree.

As an example: If the user types "Get Started", the bot will answer specifically "Hi, I am your bot"

However, after doing research on LangChain, I did not see a way to create a chatbot with a decision tree nor did I see a way to create specific responses for both the user and the bot. I am wondering if anyone can help me with this.

[chloetrujillo]

https://cobusgreyling.medium.com/langchain-is-opensource-software-for-llm-dialog-state-contextual-memory-management-bb0469fdf944 try this article!

[jsbryaniv]

I would love to see prompt decision trees implemented in LangChain. I think that would make this library very powerful. They would be useful for 1) Better predictability in outcomes of complex tasks by splitting the job into smaller easy-to-implement goals, 2) the context window would be effectively larger because instead of having one massive system message you could just focus on one smaller system message prompt at a time, and 3) Could eliminate the need for a scratch pad, which can be pretty costly if the ai writes too many notes.

[lasharela]

Hi everyone,
I wanted to build an interactive onboarding flow:

1. Would you like to register or will you stay anonymous?
2. If registering, do you prefer social auth or a one-time code?
3. If a one-time code - SMS or email, etc.

As you see, this is a decision tree. I tried to use LangGraph, but LangGraph's graph is for data processing. Every time, every invoke is going through the entire graph.

Is there any better framework or functionality which allows us to achieve this? I want to use Graph as a state. Based on users’ feedback, we have to switch nodes and save users’ state.

Anything like this?