Semantic Search Commands

Semantic Search Commands

Semantic search allows you to find documents based on meaning rather than just keyword matching. It uses vector embeddings (via ChromaDB) to understand the context of your query.

All semantic commands use crispt.

Basic Semantic Search

Find documents that are conceptually similar to your query.

crispt --inp ./corpus --semantic "patient anxiety about surgery" --num 10 --out ./search_results

• --semantic <query>: The search phrase.
• --num <N>: Number of results to return (default: 5).
• --rec <threshold>: Minimum similarity score (0.0 to 1.0). Default: 0.4.

Find Similar Documents

Find documents that are similar to a specific document (or list of documents) you already have. Great for literature reviews.

crispt --inp ./corpus --similar-docs "doc_id_1,doc_id_2" --num 5

• --similar-docs <ids>: Comma-separated list of Document IDs to use as the query.

Semantic Chunk Search

Search for specific segments (chunks) within a single document. Useful for very long transcripts.

crispt --inp ./corpus --doc-id <ID> --semantic-chunks "specific topic"

• --doc-id: Required. The ID of the document to search within.

Visualizing Embeddings

Visualize the "semantic space" of your corpus to see how documents cluster by meaning.

crispt --inp ./corpus --embedding-viz "tsne:plot.png"

• Methods: tsne, pca, umap.
• Format: method:output_filename.