Build a Customer Support Chatbot | LangGraph
In this tutorial, we create a travel assistant chatbot using LangGraph, demonstrating reusable techniques applicable to building any customer support chatbot or AI system that uses tools, supports many user journeys, or requires a high degree of control. #AI #LangGraph #llm
We start by building a simple travel assistant and progressively add complexity to better support advanced capabilities:
1. Zero-Shot Tool Executor: In the first part, we develop a simple agent with an LLM and tools, showing the limitations of this flat design for complex experiences.
2. User Confirmation: In the second part, we add user confirmation before the agent takes any sensitive actions, giving the user more control but at the cost of a less autonomous experience.
3. Conditional Interrupts: In the third part, we split tools into "safe" and "sensitive" categories, only requiring user confirmation on sensitive actions. This improves the user experience while maintaining an appropriate level of control.
4. Specialized Workflows: In the fourth part, we separate user journeys into specific "skills" or "workflows". This allows optimizing prompts and tools for each intent, leading to a more reliable and tailored user experience.
By the end of this tutorial, you'll understand key principles for designing customer support chatbots, balancing expressiveness and control to create delightful user experiences.
Chapters:
00:00 Introduction
01:15 Background: Chatbot Design Challenges
02:38 Tutorial Roadmap: From Simple to Complex
06:50 Set up Development Environment
10:04 Part 1: Designing a Simple Zero-Shot Agent
16:08 Part 2: Add User Confirmation
19:37 Part 3: Conditional Interrupts
25:10 Zero-shot Design Limitations and Solutions
27:28 Part 4: Specialized Workflows (Intro)
29:46 Workflow Design and Optimization
38:44 Testing out + Review in LangSmith
42:57 Reflecting on the Tutorial: From Simple Agent to Specialized Workflows
46:50 Conclusion and Future Directions
Additional Resources:
- Tutorial Code: [ Ссылка ]
- LangGraph Documentation: [ Ссылка ]
[ Ссылка ]
