This repository demonstrates how to build a conversational agent that answers user questions using transcripts from the Lex Fridman podcast.
The agent is built using the following Python packages: langchain, pod-gpt, pinecone-client[grpc], openai, and tqdm.
The conversational agent is built on top of a vector database that indexes transcripts from the Lex Fridman podcast using an OpenAI GPT-3.5-turbo language model. The agent retrieves answers from the vector database using a question-answering model based on the RetrievalQA class from the langchain library.
The Memory class from the langchain.chains.conversation.memory module is used to provide conversational memory to the agent. Code is based off of Pinecone examples here. Great walkthrough!
- Download a pre-built dataset of Lex Fridman podcast transcripts using datasets from Hugging Face.
- Use the pod-gpt package to create an indexer for the transcripts and index them to a Pinecone vector database.
- Initialize the RetrievalQA class from the langchain.chains module with the OpenAI GPT-3.5-turbo language model and the Pinecone vector database.
- Initialize a Tool object with the RetrievalQA object, a description of the tool, and a name for the tool.
- Initialize a Memory object from the langchain.chains.conversation.memory module.
- Use initialize_agent() from the langchain.agents module to create a conversational agent with the initialized Tool and Memory objects.
This script is open-source and licensed under the MIT License. For more details, check the LICENSE file.