Qdrant Cloud tutorial: Python vector search for RAG

This tutorial replaces a Python knowledge list and manual similarity search with Qdrant Cloud in a RAG pipeline. The speaker explains why growing collections raise concerns about search time, persistence and memory, then introduces collections and points. Each point holds an ID, an embedding vector and a payload containing the original text.

The Python walkthrough requires a Qdrant cluster endpoint and API key in an environment file, plus the Qdrant client package. It reads knowledge.txt, uses Sentence Transformers to encode its lines, and creates a collection with 384-dimensional vectors and cosine distance. The example deletes an existing collection before recreating it, so rerunning this script replaces that collection's data.

The retrieval function embeds a question and calls query_points with a limit of three and payload retrieval enabled. It passes the returned text to an LLM as context. In the demonstrated vacation query, the relevant passage states 24 days of paid leave; the other two matches are unrelated. This illustrates that requesting three results does not make all three relevant.

The setup uses a cloud database and an API-based LLM. It does not demonstrate a self-hosted or offline deployment. The speaker also warns that an unused cloud cluster may have been deleted and that learners may need fresh credentials.