Course Wiki

Keyword Search

The baseline retrieval method: matching query terms against document text with full-text search techniques.

LLM Zoomcamp starts retrieval with keyword search — no embeddings, no vector database. Documents are indexed with an inverted index (the approach behind minsearch and classic full-text engines like Elasticsearch), queries match on term overlap with optional boosting of important fields, and scoring ranks the matches.

The course’s point is practical: keyword search is cheap, explainable, and surprisingly strong on FAQ-style data. Later modules reintroduce it in hybrid search, where combining keyword and vector retrieval outperforms either alone — so the baseline is not a stepping stone to discard but a component to keep.

Taught in

• LLM Zoomcamp

  • Module 1: Agentic RAG
    • Search
    • Wrap-up of Part 1
  • Module 2: Vector Search
    • Vector Search
    • RAG with Vector Search
  • Module 4: Evaluation
    • Evaluation
  • Module 6: Best Practices
    • Hybrid Search
    • Next Steps ## Related concepts

• RAG

• Vector Search

• Hybrid Search

• Embeddings