Contextual Webinar RAG — DevOps skill for Claude Code

pc-yt-rag

A simplified Contextual Video RAG implementation using Pinecone, AWS, and Claude

Ever wanted to ask questions over your video data, such as Youtube, Zoom webinars, recorded meetings, etc? This application aims to create a RAG chatbot over these content using contextual retrieval and Pinecone, AWS, and Claude.

Wanna try it live? Click here for the [deployed application](https://pinecone-contextual-rag-demo.streamlit.app/).

This branch contains the **Streamlit Web App** version of the implementation. This allows you to run a local web app to interact with the RAG chatbot, and uses a makefile to make the data preprocessing smoother. Please read the following section to ensure you have the appropriate prerequisites before proceeding.

If you'd rather work in Sagemaker Notebook, use the webinar-notebook branch above!

Before you Begin

This repo presents the RAG solution in two ways: one using scripting and makefiles, to create a Streamlit application, and another using a notebook intended for use on Sagemaker.

You'll also need access to AWS Bedrock, Pinecone (via an API Key), and Claude specifically via Bedrock.

Finally, you need to add the videos you'd like to process under a folder called data, with a subfolder called videos. Leave them in .mp4 format. If you have access to your own Youtube channel, downloading videos from the console there will be perfect!

Running the Scripts Locally

Before beginning, authenthicate your session with AWS using your preferred method. You can save the access key, default region, and secret access key as environmental variables, or use 'aws sso login' if you have that setup.

**You'll still need access to AWS Bedrock and Claude via Bedrock, as well as a Pinecone API Key**

To run the scripts locally, you can use the provided Makefile. Below are the available commands:

1. Create the .env file:

   make create-env

   This command will create the .env file for new users and prompt you to add y