Multimodal Recipe Recommender using Qdrant, LlamaIndex, and Google Gemini 🚀

This repository contains a notebook with a multimodal system using images as frames from YouTube videos, LlamaIndex framework, Qdrant as a vector database, and Gemini as embedding and llm model.

Main Steps

• Data Ingestion: Load videos and metadata from a YouTube playlist
• Indexing: MultiModalVectorStoreIndex from LlamaIndex
• Embedding and Model: Gemini
• Vector Store: Qdrant with 2 collections (text and images)
• Query Retrieval: Top recipe and frame images

Feel free to ⭐ and clone this repo 😉

Tech Stack