Notes on Generative AI Applications with Langraph

Introduction

• Presenter: Krish Nayak
• Topic: Building generative AI applications using Langraph
• Focus: Integrating a chatbot with third-party tools
  • Example tools: Wikipedia, Google Search
• Duration: 25-30 minutes

Overview of Langraph

• Provided by Langchern for creating multi-AI agents.
• Capabilities:
  • Create agents
  • Manage state
  • Develop workflows

Application Development

Architecture Flow

1. User Query: User asks a question to the chatbot.
2. Chatbot Response: The chatbot determines:
   • If it can answer using LLMs, it provides a response.
   • If not, it queries an external tool.
3. External Tool Response: If the tool provides an answer, the chatbot replies with it. If not, the chatbot returns no answer or additional information from LLM.

Step-by-Step Implementation

1. Library Installation:

   • Install necessary libraries: langraph, langsmith, langchain, langchain_grok.
   • Use pip commands to install libraries.

2. Importing Libraries:

   • Import essential libraries for managing state and tools.
   • Example tools: Wikipedia API wrapper, R-CIF API wrapper.

3. Tool Setup:

   • Create instances for wikipedia and R-CIF wrappers.
   • Set parameters for results and character limits.

4. Creating the Langraph Application:

   • Define a state class utilizing type dict for managing messages.
   • Import necessary functions from Langraph.
   • Set up the LLM model using Grok API.
     • API keys must be stored securely.

5. Defining the Chatbot:

   • Create chatbot functionality: it will manage state and interact with LLM.
   • Integrate tools within the chatbot's processing capabilities.

6. Building the Workflow:

   • Use graph builder to create nodes and edges for the process flow:
     • Start node to Chatbot node.
     • Chatbot to Tool node.
     • Conditional edges to allow bi-directional communication between the Chatbot and tools.
     • End node to signify completion of the interaction.

7. Compiling and Testing:

   • Compile the graph and visualize it using IPython.display.
   • Execute queries to test the chatbot's functionality.

Example Test Case

• User input: "Hi there! My name is John."
  • The chatbot first tries to search for "John" in Wikipedia.
  • If no relevant result is found, it provides a default response: "Hello John, it's nice to meet you."

Conclusion

• Encouragement to experiment with multiple tools using Langraph for various queries.
• End of tutorial.
• Final words of thanks and encouragement to watch further videos.