Qdrant Vector Database - Text Chunking strategies | Python | RAG

Experimenting with 3 different text chunking strategies and 3 embedding models, using Qdrant and their Python client - for potential RAG / LLM usage. Python Qdrant AI Playlist: https://www.youtube.com/playlist?list=PLKMY3XNPiQ7vOOkkkU61jxQglxtWEX2Ef Buy me a coffee (or Microphone!) ☕ https://www.buymeacoffee.com/DrPi - chapters 00:00 intro 09:15 models and data 14:23 results 21:52 qdrant docs Models: - all-MiniLM-L6-v2 - all-MiniLM-L12-v2 - all-mpnet-base-v2 Strategies: - fixed - sentence - paragraph "fixed" - for fixed-size chunking embeddings "sentence" - for sentence-based chunking embeddings "paragraph" - for paragraph-based chunking embeddings 🔗 https://qdrant.tech/course/essentials/day-1/chunking-strategies/#1-fixed-size-chunking My code/configuration sets up three separate named vector spaces within the same collection (for each of the 3 the chunking strategies) and then 1 collection per model (So 3 collections, because we are using 3 models). Named Vectors ------------------------- Named Vectors store multiple vectors of different sizes and types in the same data point. This is useful when you need to define your data with multiple embeddings (in this example, it was different chunk strategies). 🔗 https://qdrant.tech/documentation/concepts/vectors/#named-vectors Note ! - There's no universal "best" chunking strategy - it depends on: - Content structure - Paragraph coherence - Query patterns - Information density #Qdrant #chunking #vectordatabase