Build Your own Perplexity Lite using LangGraph, GPT4, and Tavily AI

---

---

AI Summary

Video Summary: Building Perplexity Light with Lang Chain and Tavily

Introduction

• The video is about creating an application similar to Perplexity, named Perplexity Light.
• Perplexity is a startup that offers a search system competing with Google, providing features like co-pilots for human-like responses and a lab for experimenting with language models (LLMs).

Development Process

• The new Lang Chain library update (version 0.1.0) introduces Lang Graph, which allows building language agents as graphs.
• Lang Graph offers custom cyclical behavior and explicit planning, differing from the loop execution of agents in previous Lang Chain versions.
• The application will be built using Google Colab, FastAPI, and Bootstrap for the frontend.

Lang Chain and Tavily Integration

• The application will use Lang Graph from Lang Chain and Tavily, a search API that connects LLMs to the web.
• GPT-4 will be used as the LLM for the application.
• The application will be a FastAPI microservice with a frontend built using Bootstrap.

Demonstration

• The presenter demonstrates querying the application for startups in drug discovery and sustainability.
• The application provides synthesized responses with links to sources for further information.
• The presenter shows the raw trace of the application’s process, detailing intermediate steps and the final output.

Building the Application

• The presenter writes Python code in Google Colab, installing dependencies and setting up the environment.
• They define agents, import necessary modules, and create functions for executing tools and determining if the process should continue.
• They demonstrate how to use Lang Graph to build the application’s workflow with nodes and conditional edges.
• The application is then compiled and executed, with the presenter showing the output.

FastAPI and Frontend Setup

• The presenter explains the frontend HTML structure, including input boxes, loaders, and answer cards.
• They show the JavaScript functions used to fetch and display the API’s responses.
• In app.py, they set up FastAPI, mount directories, and create endpoints.
• The main function processes queries and invokes the chain to get responses.

Conclusion

• The presenter runs the FastAPI application, demonstrating its functionality in a web browser.
• They encourage viewers to experiment with the code and provide feedback.

Additional Notes

• The presenter encountered some issues with module imports and JSON serialization, which they addressed in the code.
• They mention the potential for JSON errors and leave it to the viewers to handle such cases.

Call to Action

• The presenter invites feedback, encourages likes and subscriptions, and offers the code for viewers to use.