Check if text contains financial figures ($, %, large numbers).

Perform hybrid search combining BM25 and vector search. Args: query: Search query k: Number of final results bm2

Perform hybrid search combining BM25 and vector search.

Run interactive Q&A with graph-enhanced retrieval.

Run interactive Q&A with hybrid search.

Run an interactive Q&A session with filter support.

Run an interactive Q&A session.

Run interactive Q&A with the agent.

Judge the quality of an answer.

Use LLM to judge the quality of decomposition.

Use LLM to judge if an answer is grounded in the context.

Use LLM to judge if a document is relevant to the question.

Use LLM to judge if a document is relevant to the question.

Load PDFs and split into chunks.

Load PDFs and split into chunks with rich metadata.

Load document chunks from MongoDB.

Test graph-enhanced retrieval.

Run example queries or interactive mode.

Build the knowledge graph from MongoDB chunks.

Run the entity extraction evaluation.

Run the multi-hop reasoning evaluation.

Test hybrid retrieval and compare with individual methods.

Run example queries or interactive mode.

Test retrieval with various filters.

Run example queries or interactive mode.

Main ingestion pipeline with fast metadata extraction.

Run the latency evaluation.

Run the precision delta evaluation.

Test retrieval with a sample query.

Run example queries or interactive mode.

Main ingestion pipeline.

Run the groundedness evaluation.

Run the precision evaluation.

Test the agentic RAG system.

Run the interactive agent.

Run the query decomposition evaluation.

Run the tool selection evaluation.

Run the end-to-end evaluation.

Measure generation latency and return answer. Returns: Tuple of (answer, latency_ms)

Measure retrieval latency and return documents. Returns: Tuple of (documents, latency_ms)

Print statistics about the graph.

Print summary of metadata in the collection.

Print instructions for creating the vector search index in MongoDB Atlas.

Print instructions for creating the vector search index in MongoDB Atlas.

Combine ranked lists using Reciprocal Rank Fusion.

Combine multiple ranked lists using Reciprocal Rank Fusion. RRF Score = sum(weight_i / (k + rank_i)) for each list Args:

Combine ranked lists using Reciprocal Rank Fusion.

Retrieve using all three methods and compare results.

Retrieve the top-k most relevant documents for a given query. Args: query: The search query string top_k: Number of docu

Retrieve top-k documents for a query.

Retrieve top-k documents for a query.

Retrieve the top-k most relevant documents with similarity scores. Args: query: The search query string top_k: Number of

Run entity extraction evaluation on all test cases.

Run multi-hop reasoning evaluation on all test cases.

Run precision delta evaluation on all test cases. Returns: Dictionary with aggregate results

Run groundedness evaluation on all test cases. Returns aggregate results and per-question breakdown.

Run precision evaluation on all test cases. Returns aggregate results and per-question breakdown.

Run query decomposition evaluation on all test cases.

Run tool selection evaluation on all test cases.

Run end-to-end evaluation on all test cases.

Run latency evaluation across all test cases. Args: test_cases: List of test cases to run k: Number of documents to retr

Run a single latency test. Returns: Dictionary with latency measurements

Set up MongoDB collection for vector storage.

Set up MongoDB collection for vector storage.

Create indexes and constraints in Neo4j.

Synthesize final answer from all retrieved contexts.

Standard vector similarity search. Args: query: Search query k: Number of documents to retrieve Returns:

Find paths between two entities in the graph.

Format retrieved documents into context string for LLM.

Judge if an answer is grounded in the context.

Retrieve documents with metadata pre-filtering and similarity scores. Returns: List of tuples (document, score)