Skip to content
Prompt Engineering with GPT & LangChain
  • AI Chat
  • Code
  • Report

    • Prompt Engineering with GPT and LangChain

    LangChain is framework that is extremely helpful for prompt engineering and the integration of generative AI capabilities in applications or data platforms. It has many capabilities, some of which will not be introduced until later modules, but we will start with a gentle introduction to some of the easy-to-understand concepts in the framework.

    You'll build an AI agent that uses Python and GPT to perform sentiment analysis on financial headlines.

    In more detail, you'll cover:

    • Getting set up with an OpenAI developer account and integration with Workspace.
    • Interacting with OpenAI models through the langchain framework.
    • Using prompt templates that write reusable, dynamic prompts.
    • Working with LLM chains.
    • Automatically parsing the output of an LLM to be used downstream.
    • Working with langchain agents and tools.
    • Using the OpenAI Moderation API to filter explicit content.

    For this project, we are using two small samples: financial_headlines.txt and reddit_comments.txt. These 5-6 line samples are kept short to keep evaluation easy, but keep in mind that this same code and prompt engineering techniques can scale to datasets of much larger size.

    Before you begin

    You'll need a developer account with OpenAI.

    See getting-started.ipynb for steps on how to create an API key and store it in Workspace. In particular, you'll need to follow the instructions in the "Getting started with OpenAI" and "Setting up Workspace Integrations" sections.

    Task 0: Setup

    We need to install a few packages, one of which being the langchain package. This is currently being developed quickly, sometimes with breaking changes, so we fix the version.

    langchain depends on a recent version of typing_extensions, so we need to update that package, again fixing the version.

    Instructions

    Run the following code to install openai, langchain, typing_extensions and pandas.

    # Install openai.
!pip install openai==0.28.0
    Hidden output 
    # Install langchain.
!pip install langchain==0.0.293
    Hidden output 
    # Install typing-extensions.
!pip install typing-extensions==4.8.0
    Hidden output 
    # Install pandas.
!pip install pandas==2.0.3
    Hidden output

    For this project, we need first need to load the openai and os packages to set the API key from the environment variables you just created.

    Instructions

    • Import the os package.
    • Import the openai package.
    • Set openai.api_key to the OPENAI_API_KEY environment variable.
    # Import the os package.
import os

# Import the openai package.
import openai

# Set openai.api_key to the OPENAI_API_KEY environment variable.
# Use os.getenv to avoid KeyError if the environment variable is not set.
openai.api_key = os.getenv("OPENAI_API_KEY")

    For the langchain package, let's start by importing it's OpenAI and ChatOpenAI class, which are used to interact with completion models and chat completion models respectively.

    Completion models, such as the GPT-1, GPT-2, GPT-3 and GPT-3.5, work as an advanced autocomplete model. Given a certain snippet of text as input, they will complete the text until a certain point. This could be either an end-of-sequence token (a natural way of stopping), the model reaching its maximum token limit for outputs and so on.

    Chat completion models, such as GPT-3.5-Turbo (the ChatGPT model) and GPT-4, are designed for conversational use. These models are typically more fine-tuned for conversations, keep a prompt/conversation history and allow access to a system message, which we can use as a meta prompt to define a role, a tone of voice, a scope, etc.

    Completion models and chat completion models tend to work with different classes and functions in the SDK. For that reason, we will start by importing both classes.

    ‌
    ‌
    ‌