Experiment using Langchain, OpenAI and Streamlit, along with FAISS for CPU vector store, that analyzes YouTube transcripts and answers questions

01_setup.sh

#!/usr/bin/env bash

# reference: https://www.youtube.com/watch?v=lG7Uxts9SXs    (some code was deprecated so fix along way)

# create folder
mkdir youtube-assistant
cd youtube-assistant

# set up virtual env
python3 -m venv .venv
source .venv/bin/activate

# install dependencies
pip3 install -U python-dotenv langchain langchain-community langchain-openai openai youtube-transcript-api streamlit faiss-cpu

# create .env
echo "OPENAI_API_KEY=\"sk-YOURKEYHERE\"" > .env

# test the helper
python3 langchain_helper.py  # prints object ID (hoping for no errors, warnings expected)

# run the chat bot using streamlit
streamlit run main.py

langchain_helper.py

from langchain import PromptTemplate
from langchain.chains import LLMChain
from langchain_community.document_loaders import YoutubeLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAI
from langchain_openai import OpenAIEmbeddings

from dotenv import load_dotenv

load_dotenv()

embeddings = OpenAIEmbeddings()

def create_vector_db_from_youtube_url(video_url: str) -> FAISS:
    loader = YoutubeLoader.from_youtube_url(video_url)
    transcript = loader.load()

    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
    docs = text_splitter.split_documents(transcript)

    db = FAISS.from_documents(docs, embeddings)
    return db

def get_response_from_query(db, query, k=4):
    # text-davinci can handle 4097 tokens
    docs = db.similarity_search(query, k)
    docs_page_content = " ".join([d.page_content for d in docs])

    llm = OpenAI(model="gpt-3.5-turbo-instruct") # https://platform.openai.com/docs/deprecations

    prompt = PromptTemplate(
        input_variables = ["question", "docs"],
        template = """
        You are a helpful YouTube assistant that can answer questions about videos based on the video's transcript.

        Answer the following question: {question} 
        By searching the following video transcript: {docs}

        Only use the factual information from the transcript to answer the question.

        If you feel like you don't have enough information to answer the question, say "I don't know".

        Your answers should be detailed.
        """
    )

    chain = LLMChain(llm = llm, prompt = prompt)

    response = chain.run(question = query, docs = docs_page_content)
    response = response.replace("\n", "")
    return response, docs


if __name__ == "__main__":
    print("Testing ...")
    video_url = "https://youtu.be/-Osca2Zax4Y?si=iyOiePxzUy_bUayO"
    result = create_vector_db_from_youtube_url(video_url)
    print(result) # just prints object ID to confirm it created it

main.py

import langchain_helper as lch
import streamlit as st
import textwrap

st.title("YouTube Assistant")

with st.sidebar:
    with st.form(key='my_form'):
        youtube_url = st.sidebar.text_area(
            label="What is the YouTube video URL?",
            max_chars=50
        )
        query = st.sidebar.text_area(
            label="Ask me about the video?",
            max_chars=50,
            key="query"
        )

        submit_button = st.form_submit_button(label="Submit")

if query and youtube_url:
    db = lch.create_vector_db_from_youtube_url(youtube_url)
    response, docs = lch.get_response_from_query(db, query)
    st.subheader("Answer:")
    st.text(textwrap.fill(response, width = 80))

Generative AI (GenAI) YouTube Video Helper

This example from second part of a Langchain Tutorial teaches you how to transcribe text from YouTube videos, then embed and store the chunked text into a vector store using FAISS, and ask questions about the video using only the transcribed text.

Result

Valid query

Invalid query

Test with different video on Kong API Gateways