Build the PRA Knowledge Base for RAG Retrieval — build_knowledge

Reads the curated risk analysis knowledge files bundled with the PRA package, chunks them, generates embeddings via Ollama, and stores them in a DuckDB-backed ragnar knowledge base for retrieval-augmented generation.

Usage

build_knowledge_base(
  store_path = NULL,
  embed_model = "nomic-embed-text",
  overwrite = FALSE
)

Arguments

store_path: Path to store the DuckDB knowledge base. Defaults to a cache directory under tools::R_user_dir().
embed_model: Ollama embedding model name (default "nomic-embed-text").
overwrite: Logical. If TRUE, rebuild the knowledge base even if a cached version exists. Default FALSE.

Value

A ragnar store object that can be passed to retrieve_context().

Details

The knowledge base is built once and cached to disk. Subsequent calls with the same store_path load the existing store.

Examples

if (FALSE) { # \dontrun{
store <- build_knowledge_base()
context <- retrieve_context(store, "How do I run a Monte Carlo simulation?")
} # }