Log message if verbose mode is on.

Run the agentic RAG pipeline. Returns: Dictionary with answer and metadata about the agent's decisions

Retrieve documents with metadata pre-filtering. Args: query: The search query string top_k: Number of documents to retri

Execute a retrieval tool.

Evaluate if retrieved results are sufficient.

Check if two entities match (allowing for variations).

Generate an answer using graph-enhanced RAG. Args: question: User's question top_k: Number of documents to retrieve

Generate an answer using hybrid RAG.

Generate an answer using the metadata-filtered RAG pipeline. Args: question: The user's question top_k: Number of docume

Generate an answer using the naive RAG pipeline. Args: question: The user's question top_k: Number of documents to retri

Get MongoDB client.

Run the agent's query analyzer on a question.

Run the agent's query analyzer on a question.

Analyze the query to determine retrieval strategy.

Build a human-readable description of the applied filters.

Build a MongoDB pre-filter for vector search. Args: year: Single year or list of years to include year_range: Tuple of (

Calculate precision for a set of retrieved documents. Returns: Tuple of (precision_score, list_of_relevance_judgments)

Calculate precision for a single query. Precision = relevant_documents / k Returns dict with precision score and details.

Run full end-to-end evaluation for a single test case.

Evaluate groundedness for a single question. Returns dict with groundedness assessment and details.

Extract entities from a question using the LLM.

Find entities related via graph traversal.

Format documents into context string for LLM.

Format retrieved documents into a context string for the LLM. Args: documents: List of retrieved document objects Retur

Get counts of company mentions in the collection.

Get MongoDB client.

Get MongoDB client.

Get counts of each topic bucket in the collection.

Connect to the MongoDB vector store.

Connect to the MongoDB vector store.

Connect to the MongoDB vector store.

Connect to MongoDB vector store.

Perform graph-enhanced retrieval. Steps: 1. Extract entities from query 2. Traverse graph to find related entities 3. Get ch

Use LLM to judge if the traversal path is relevant.

Load documents from MongoDB for BM25 indexing. Note: BM25 requires all documents in memory. For very large collections, consider usi

Load documents from MongoDB for BM25 indexing.

Skip retrieval - agent decides to answer from model knowledge. Returns: Empty result indicating no retrieval was done

Normalize entity name for comparison.

Retrieve documents with optional filters.

Perform vector similarity search.

Add high-level topic nodes based on entity patterns.

Build a description of the entities found.

Build the knowledge graph from extracted entities and relationships.

Build a MongoDB pre-filter for vector search.

Build a MongoDB pre-filter for vector search.

Build MongoDB pre-filter.

Calculate precision, recall, and F1 for entity extraction.

Clear all nodes and relationships from the graph.

Create LLM chain for entity/relationship extraction.

Create the LLM judge for assessing groundedness.

Create the RAG chain.

Create the RAG chain.

Create the RAG chain with prompt template and LLM.

Create the RAG chain with prompt template and LLM.

Create the LLM judge for assessing relevance.

Create the LLM judge for assessing relevance.

Create embeddings and store in MongoDB Atlas Vector Search.

Create embeddings and store in MongoDB Atlas Vector Search.

Check collection status.

Debug function to check MongoDB collection status.

Debug function to check MongoDB collection status.

Detect mentions of Berkshire portfolio companies.

Detect mentions of Berkshire wholly-owned subsidiaries.

Detect which topic buckets apply to this text using keyword matching.

Evaluate query decomposition for a single test case.

Evaluate entity extraction for a single test case.

Evaluate multi-hop reasoning for a single test case.

Compare precision with and without filters for a single question. Returns: Dictionary with precision comparison results

Evaluate tool selection for a single test case.

Extract entities and relationships from a chunk.

Extract all metadata for a chunk using fast string matching.

Extract entities from the user's question using LLM.

Extract entities from a question using the LLM.

Vector search with metadata pre-filtering. Args: query: Search query k: Number of documents year: Filter by spec

Find document names that mention any of the given entities.

Find entities related to the query entities via graph traversal. Returns: Dictionary mapping entity names to their hop distance

Format documents into context string.

Format documents into context string.

Format retrieved documents into context string.

Format retrieved documents into context string.

Format retrieved documents into a context string for the LLM.

Generate an answer using the RAG pipeline.

Get list of years available in the collection.

Get chunks from MongoDB by source file names.

Get MongoDB client.

Get MongoDB client.

Get all PDF files from the letters directory.

Get all PDF files from the letters directory.

Connect to MongoDB vector store.

Connect to the MongoDB vector store.

Connect to the MongoDB vector store.

Connect to the MongoDB vector store.

Connect to the MongoDB vector store.