from __future__ import annotations import logging from legora.config import get_settings from legora.logging import log_stage from retrieval.hybrid_retrieval import HybridRetriever from retrieval.judgments import KanoonClient logger = logging.getLogger(__name__) class LegalRetriever: def __init__(self): self.hybrid = HybridRetriever() self.kanoon = KanoonClient() # ===================================================== # BUILD CONTEXT # ===================================================== def build_context( self, hybrid_result, judgments, include_full_docs: bool = False ): context = [] # ----------------------------------------- # Statutory Retrieval # ----------------------------------------- context.append("===== STATUTORY LAW =====\n") for point in hybrid_result["reranked_results"]: payload = point.payload if include_full_docs: doc_text = payload.get('document', '') else: doc_text = payload.get('text', '') context.append( f""" DOCUMENT: {payload.get('document')} CHUNK: {payload.get('chunk_id')} TEXT: {doc_text} """ ) # ----------------------------------------- # Graph Expansion # ----------------------------------------- context.append("\n===== KNOWLEDGE GRAPH =====\n") for node_id, graph_data in hybrid_result["graph_context"].items(): context.append(f"\nNODE: {node_id}") context.append( f""" Ancestors: {graph_data['ancestors']} Children: {graph_data['children']} References: {graph_data['references']} """ ) # ----------------------------------------- # Judgments # ----------------------------------------- context.append("\n===== JUDGMENTS =====\n") for judgment in judgments: context.append(str(judgment)) return "\n".join(context) # ===================================================== # RETRIEVE # ===================================================== def retrieve( self, query: str, **kwargs ): settings = get_settings() # Extract known parameters with defaults vector_k = kwargs.get('vector_k', settings.top_k) rerank_k = kwargs.get('rerank_k', settings.rerank_k) judgment_k = kwargs.get('judgment_k', settings.judgment_k) include_full_docs = kwargs.get('include_full_docs', False) # Allow 'top_k' as alias for vector_k if 'top_k' in kwargs: vector_k = kwargs['top_k'] # Forward only options supported by each retriever. graph_options = {} if kwargs.get("include_graph", True) else {"include_graph": False} hybrid_result = self.hybrid.retrieve( query=query, vector_k=vector_k, rerank_k=rerank_k, **graph_options ) judgments = [] if kwargs.get("include_judgments", True): with log_stage("judgments") as metrics: judgments = self.kanoon.retrieve( query=query, top_k=judgment_k ) metrics.candidate_count = len(judgments) final_context = self.build_context( hybrid_result, judgments, include_full_docs=include_full_docs ) return { "query": query, "statutory_results": hybrid_result["reranked_results"], "vector_results": hybrid_result.get("vector_results", []), "node_ids": hybrid_result.get("node_ids", []), "graph_context": hybrid_result["graph_context"], "judgments": judgments, "final_context": final_context } if __name__ == "__main__": query = "My bike was stolen from outside my house. What punishment can the offender face?" retriever = LegalRetriever() logger.debug('%s', retriever.retrieve(query))