AI Tool Weaviate Hybrid Search: Combining Keyword and Vector Recall

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-tools-weaviate-hybrid-search-r9a4-creators

1. In Weaviate hybrid search, what two ranking methods are combined?

   1. BM25 keyword ranking and dense vector similarity scoring
   2. TF-IDF weighting and sparse vector embeddings
   3. Inverted index and forward index
   4. Lexical matching and semantic clustering

2. What does the alpha parameter control in Weaviate hybrid search?

   1. The maximum number of results to return
   2. The balance between BM25 and vector ranking contributions to final scores
   3. The dimensionality of the embedding vectors
   4. The threshold for accepting search results

3. Which task can AI reliably perform when working with Weaviate hybrid search?

   1. Scaffolding query code with metadata filters and per-class configuration
   2. Automatically detecting if BM25 tokenization matches your domain
   3. Choosing which search algorithm to use based on data characteristics
   4. Determining the perfect alpha value for any dataset

4. Why does AI need labeled ground truth data to determine the optimal alpha?

   1. Labeled data is required to generate the initial BM25 index
   2. Alpha is a random parameter that only works with synthetic test data
   3. Vector models cannot run without pre-labeled query-result pairs
   4. Alpha must be tuned against known relevant results to maximize actual recall precision

5. What is required to properly tune the alpha parameter in Weaviate hybrid search?

   1. A minimum of 10,000 indexed documents
   2. A GPU-accelerated compute environment
   3. A pre-trained language model fine-tuned on your data
   4. A labeled set of query-document pairs with known relevance

6. What type of notebook can AI help produce for Weaviate hybrid search optimization?

   1. A notebook that automatically retrains vector embeddings nightly
   2. A visualization dashboard for monitoring BM25 index health
   3. A natural language interface for querying the database without code
   4. An alpha sweep notebook that tests multiple alpha values against labeled data

7. What typically happens when you use default alpha values for specialized corpora?

   1. The search automatically switches to pure BM25 mode
   2. The vector index is disabled to prioritize keyword results
   3. The search recall is usually suboptimal because defaults don't fit specialized data
   4. Default alpha values work perfectly for any domain

8. In the alpha sweep process, what does 'sweeping' alpha mean?

   1. Re-indexing documents with different vector models
   2. Adjusting the weighting between query terms
   3. Removing outliers from the BM25 scoring function
   4. Systematically testing many alpha values to find the best-performing one

9. What is a fallback plan needed for in Weaviate hybrid search deployments?

   1. An alternative to using alpha parameters
   2. A backup method if BM25 returns zero results
   3. A strategy to use when the vector index becomes unavailable
   4. A plan to switch between different embedding models

10. What does 'scaffolding' mean when AI generates Weaviate hybrid search code?

    1. AI replaces the need for any configuration files
    2. AI creates the structural framework and code that humans then refine and tune
    3. AI fully implements and deploys the search system without human input
    4. AI automatically optimizes all parameters before deployment

11. What does per-class configuration allow in Weaviate hybrid search?

    1. Automatic classification of queries into BM25 or vector categories
    2. Selective encryption of specific document types
    3. Different alpha values and ranking behaviors for different data classes or collections
    4. Faster query processing by skipping certain document classes

12. Why must alpha be swept with a labeled set per index rather than once globally?

    1. Global alpha would cause security vulnerabilities
    2. Alpha values cannot be transferred between indexes due to licensing
    3. Each index has different relevance patterns requiring independent optimization
    4. Labeled sets from one index are incompatible with others

13. What role do metadata filters play in hybrid search queries that AI can generate?

    1. They replace BM25 when filters are applied
    2. They restrict results to documents matching specific metadata criteria
    3. They automatically generate keywords from document metadata
    4. They increase the vector embedding dimension for metadata fields

14. What is the fundamental limitation preventing AI from choosing the right alpha automatically?

    1. AI lacks access to domain-specific relevance judgments that humans possess
    2. AI cannot read the Weaviate documentation
    3. Alpha values are randomly generated and unpredictable
    4. Vector search and BM25 are incompatible in AI systems

15. What should happen after AI scaffolds a Weaviate hybrid search query?

    1. The query is ready for production deployment without changes
    2. The vector index should be disabled to verify BM25 performance
    3. AI will automatically continue optimizing the query over time
    4. The search team should tune alpha and verify tokenization with domain expertise