Agentic RAG Playground

Splits policy documents into semantically meaningful chunks with overlap for context preservation.

Converts text into dense vector representations for semantic similarity matching.

Retrieves the most semantically relevant chunks from the vector store using approximate nearest neighbor search.

Cross-encoder reranking model that scores query-document pairs for fine-grained relevance.

Generates multiple query reformulations to improve recall and catch documents with different terminology.

Grounds every claim in the response to a specific source document with page numbers and confidence scores.