Webinar Vibe Coding Rag

---

title: "Qdrant Webinar: Vibe Coding Rag" emoji: 🏆 colorFrom: pink colorTo: indigo sdk: docker pinned: false license: apache-2.0 short_description: YouTube In-Video Search

webinar-vibe-coding-rag

[](https://huggingface.co/spaces/Qdrant/webinar-vibe-coding-rag)

**This repository contains materials for the hands-on "[Letting LLMs Write RAG Applications](https://try.qdrant.tech/llm-rag)" webinar.**

Project Overview: YouTube In-Video Search

When learning a new skill, YouTube videos can be a great resource. However, in-depth content is often lengthy and may assume no prior knowledge. What if you could have a smart assistant to help you navigate through videos and find exactly what you need? This project creates a search engine for video content, helping you skim through and focus on what matters specifically to you.

Retrieval Augmented Generation (RAG) is perfect for this task. By indexing a video's transcript, we provide an interface to search through its content. Users can click on search results to jump to the exact timestamp where a topic is discussed.

How It Works

The application has two main views:

1. **Input View**:

   • User provides a YouTube video URL
   • Backend processes the video by:
     • Extracting the transcript
     • Dividing it into 30-second chunks with 10-second overlaps
     • Creating embeddings using SentenceTransformers
     • Storing these vectors in Qdrant

2. **Video View**:

   • Displays the video alongside its transcription
   • Allows clicking on timestamps to jump to specific parts
   • Provides a search bar to filter content
   • When a query is submitted, backend returns the most relevant video segments
   • Results appear as clickable links, while hiding irrelevant parts of the transcript

If a video has been processed previously, the application uses existing embeddings without reprocessing.

Technologies