# Copyright (C) 2026 Mattia Zatelli # # This program is free software: you can redistribute it and/or modify # it under the terms of the GNU Affero General Public License as published by # the Free Software Foundation, either version 3 of the License, or # (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU Affero General Public License for more details. # # You should have received a copy of the GNU Affero General Public License # along with this program. If not, see . """ Query model for storing natural language queries and results This model manages the entire lifecycle of BIM queries: - Storage of original and processed queries - NLP/LLM processing metadata - Vector embeddings for semantic search - Complex multi-intent query support - Result caching for performance - Analytics and feedback for ML - Billing system integration Supports both simple and complex queries with automatic decomposition and dependency management between sub-queries. """ import enum from typing import Dict, Any, List, Optional from sqlalchemy import Column, Integer, String, Text, JSON, Float, Boolean, ForeignKey, DateTime from sqlalchemy.orm import relationship from sqlalchemy.sql import func from pgvector.sqlalchemy import Vector from app.core.database import Base class QueryType(str, enum.Enum): """Query types supported by the system - NATURAL_LANGUAGE: Natural language queries (Italian/English) - STRUCTURED: Structured programmatic queries - BATCH: Batch processing of multiple queries - CALCULATE: Specific calculations (area, volume, counts) - EXPORT: Data export requests """ NATURAL_LANGUAGE = "nl" STRUCTURED = "structured" BATCH = "batch" CALCULATE = "calculate" EXPORT = "export" class QueryStatus(str, enum.Enum): """Query processing states - PENDING: Query received, waiting for processing - PROCESSING: Processing in progress (NLP, database, calculations) - COMPLETED: Processing completed successfully - FAILED: Processing failed (parsing error, timeout, etc.) - CACHED: Result retrieved from cache (no processing) """ PENDING = "pending" PROCESSING = "processing" COMPLETED = "completed" FAILED = "failed" CACHED = "cached" class Query(Base): """ Query model for storing NL queries and results **Identification and Security:** - Link to project with CASCADE foreign key - Mandatory tenant_id isolation - Query hash for deduplication and cache **NLP/LLM Processing:** - Original and processed query - Vector embeddings for semantic search (1536 dim) - LLM metadata (model, tokens, reasoning) - Classification time and performance **Multi-Intent Support:** - Support for decomposed complex queries - Sub-queries with relationships and dependencies - Primary intent for backward compatibility **Results and Cache:** - Structured JSON result data - Human-readable explanations generated by LLM - Provenance for traceability - Cache with confidence score **Analytics and ML:** - User feedback for training - Performance metrics for optimization - Query patterns for insights Used for: - Semantic search for similar queries - Intelligent result caching - Usage pattern analytics - Custom NLP model training - Billing integration """ __tablename__ = "queries" # === PRIMARY IDENTIFICATION === id = Column(Integer, primary_key=True, index=True) # === RELATIONSHIPS AND SECURITY === project_id = Column(Integer, ForeignKey("projects.id", ondelete="CASCADE"), nullable=False, index=True) tenant_id = Column(String(50), nullable=False, index=True) # CRITICAL: tenant isolation api_key_hash = Column(String(100), nullable=True, index=True) # For billing and analytics # === QUERY INFORMATION === query_type = Column(String(20), nullable=False, default=QueryType.NATURAL_LANGUAGE.value) original_query = Column(Text, nullable=False) # Original user query processed_query = Column(Text, nullable=True) # Cleaned/normalized query query_hash = Column(String(64), nullable=False, index=True) # SHA256 for cache/deduplication # Plain text prompt without metadata plain_text_prompt = Column(Text, nullable=True) # Vector embedding for semantic search (OpenAI ada-002: 1536 dimensions) query_embedding = Column(Vector(1536), nullable=True) # === PROCESSING STATE === status = Column(String(20), nullable=False, default=QueryStatus.PENDING.value) processing_time_ms = Column(Integer, nullable=True) # Total processing time error_message = Column(Text, nullable=True) # Human-readable error message # === MULTI-INTENT SUPPORT === # For complex queries that require multiple coordinated processing steps is_multi_intent = Column(Boolean, nullable=False, default=False) sub_queries = Column(JSON, nullable=True) # Sub-query plan array intent_relationships = Column(JSON, nullable=True) # Relationships between sub-queries # Primary intent (for backward compatibility with existing system) primary_intent = Column(String(50), nullable=True, index=True) primary_subintent = Column(String(100), nullable=True, index=True) # === LLM PROCESSING METADATA === # Hint for result presentation format presentation_hint = Column(String(20), nullable=False, default="text") # LLM model information used llm_model_used = Column(String(50), nullable=True, index=True) # e.g. "gpt-4o-mini" tokens_used = Column(Integer, nullable=True) # Tokens consumed classification_time_ms = Column(Integer, nullable=True) # Classification time only classification_reasoning = Column(Text, nullable=True) # LLM reasoning # Legacy query structure (kept for backwards compatibility) intent = Column(String(50), nullable=True) # count, calculate, list, find, etc. entities = Column(JSON, nullable=True) # Extracted entities filters = Column(JSON, nullable=True) # Applied filters metrics = Column(JSON, nullable=True) # Requested metrics aggregations = Column(JSON, nullable=True) # Aggregation functions # Generated SQL/operations generated_sql = Column(Text, nullable=True) executed_operations = Column(JSON, nullable=True) # List of operations performed # Results result_data = Column(JSON, nullable=True) # Raw result data result_summary = Column(Text, nullable=True) # Human-readable summary result_explanation = Column(Text, nullable=True) # AI-generated explanation result_count = Column(Integer, nullable=True) # Number of elements/results result_units = Column(String(20), nullable=True) # Units (sqm, m3, count, etc.) # Enhanced result metadata result_provenance = Column(JSON, nullable=True) # Element IDs that contributed to result data_quality_notes = Column(JSON, nullable=True) # Data quality observations suggested_actions = Column(JSON, nullable=True) # Suggested follow-up actions related_queries = Column(JSON, nullable=True) # Suggested related queries # Processing metadata elements_processed = Column(Integer, nullable=True) # Number of elements analyzed total_processing_time_ms = Column(Integer, nullable=True) # Total time including sub-queries # Confidence and quality metrics confidence_score = Column(Float, nullable=True) # 0.0 - 1.0 accuracy_score = Column(Float, nullable=True) # If validated complexity_score = Column(Float, nullable=True) # Query complexity # Context and options query_options = Column(JSON, nullable=True) # Query-specific options context_data = Column(JSON, nullable=True) # Additional context used # Caching and performance cache_key = Column(String(100), nullable=True, index=True) cache_hit = Column(Boolean, default=False) similar_query_id = Column(Integer, ForeignKey("queries.id"), nullable=True) # Export and sharing is_exportable = Column(Boolean, default=True) export_formats = Column(JSON, nullable=True) # Available export formats shared = Column(Boolean, default=False) # Feedback and learning user_feedback = Column(Integer, nullable=True) # 1-5 rating feedback_comments = Column(Text, nullable=True) marked_as_favorite = Column(Boolean, default=False) # Usage tracking view_count = Column(Integer, default=0) last_accessed = Column(DateTime(timezone=True), nullable=True) # Timestamps created_at = Column(DateTime(timezone=True), server_default=func.now()) updated_at = Column(DateTime(timezone=True), server_default=func.now(), onupdate=func.now()) completed_at = Column(DateTime(timezone=True), nullable=True) # Relationships project = relationship("Project", back_populates="queries") similar_query = relationship("Query", remote_side=[id]) usage_events = relationship("UsageEvent", back_populates="query") results = relationship("QueryResult", back_populates="query", cascade="all, delete-orphan") sub_query_results = relationship("SubQueryResult", back_populates="query", cascade="all, delete-orphan") def __repr__(self): return f"" def to_dict(self) -> Dict[str, Any]: """Convert to dictionary for API responses""" return { "id": f"qry_{self.id}", "project_id": f"prj_{self.project_id}", "query_type": self.query_type, "original_query": self.original_query, "status": self.status, "processing_time_ms": self.processing_time_ms, "intent": self.intent, "entities": self.entities or [], "filters": self.filters or {}, "metrics": self.metrics or [], "result_summary": self.result_summary, "result_explanation": self.result_explanation, "result_count": self.result_count, "result_units": self.result_units, "confidence_score": self.confidence_score, "is_exportable": self.is_exportable, "export_formats": self.export_formats or [], "user_feedback": self.user_feedback, "marked_as_favorite": self.marked_as_favorite, "created_at": self.created_at.isoformat() if self.created_at else None, "completed_at": self.completed_at.isoformat() if self.completed_at else None, } def get_full_result(self) -> Dict[str, Any]: """Get complete result including data for API responses""" base_dict = self.to_dict() base_dict.update({ "result_data": self.result_data, "filters": self.filters, "aggregations": self.aggregations, "context_data": self.context_data, "query_options": self.query_options, }) return base_dict @classmethod def get_public_id(cls, internal_id: int) -> str: """Convert internal ID to public format""" return f"qry_{internal_id}" @classmethod def parse_public_id(cls, public_id: str) -> Optional[int]: """Parse public ID to internal ID""" if not public_id.startswith("qry_"): return None try: return int(public_id[4:]) except ValueError: return None def mark_completed(self, processing_time_ms: int = None): """Mark query as completed""" self.status = QueryStatus.COMPLETED.value self.completed_at = func.now() if processing_time_ms: self.processing_time_ms = processing_time_ms def mark_failed(self, error_message: str): """Mark query as failed""" self.status = QueryStatus.FAILED.value self.error_message = error_message self.completed_at = func.now() def mark_cached(self, similar_query_id: int = None): """Mark query as served from cache""" self.status = QueryStatus.CACHED.value self.cache_hit = True if similar_query_id: self.similar_query_id = similar_query_id self.completed_at = func.now() def increment_view_count(self): """Increment view count and update last accessed""" self.view_count += 1 self.last_accessed = func.now() def set_user_feedback(self, rating: int, comments: str = None): """Set user feedback for query""" if 1 <= rating <= 5: self.user_feedback = rating self.feedback_comments = comments def calculate_complexity_score(self) -> float: """Calculate query complexity score based on various factors""" score = 0.0 # Base complexity from query length query_length = len(self.original_query.split()) if self.original_query else 0 score += min(query_length / 20, 0.3) # Max 0.3 for length # Entity complexity if self.entities: entity_count = len(self.entities) score += min(entity_count / 10, 0.2) # Max 0.2 for entities # Filter complexity if self.filters: filter_count = len(self.filters) score += min(filter_count / 5, 0.2) # Max 0.2 for filters # Aggregation complexity if self.aggregations: agg_count = len(self.aggregations) score += min(agg_count / 3, 0.2) # Max 0.2 for aggregations # Processing time factor if self.processing_time_ms: time_factor = min(self.processing_time_ms / 10000, 0.1) # Max 0.1 for time score += time_factor return min(score, 1.0) def get_similar_queries(self, limit: int = 5) -> List[Dict]: """Get similar queries based on embedding similarity""" # This would use vector similarity search # Placeholder for now return [] def is_successful(self) -> bool: """Check if query was successful""" return self.status in [QueryStatus.COMPLETED.value, QueryStatus.CACHED.value] def has_results(self) -> bool: """Check if query has results""" return bool(self.result_data) or bool(self.result_summary) def get_metrics_summary(self) -> Dict[str, Any]: """Get summary of query metrics for analytics""" return { "query_id": f"qry_{self.id}", "query_type": self.query_type, "status": self.status, "processing_time_ms": self.processing_time_ms, "confidence_score": self.confidence_score, "complexity_score": self.complexity_score, "result_count": self.result_count, "cache_hit": self.cache_hit, "user_feedback": self.user_feedback, "view_count": self.view_count, "created_at": self.created_at.isoformat() if self.created_at else None, } # Multi-intent helper methods def set_multi_intent_data(self, query_plan, classification_result): """Set multi-intent data from query plan and classification result""" self.is_multi_intent = query_plan.is_multi_intent self.sub_queries = [sq.dict() for sq in query_plan.sub_queries] if query_plan.sub_queries else [] self.intent_relationships = query_plan.intent_relationships self.primary_intent = query_plan.primary_intent.value if query_plan.primary_intent else None # Set primary subintent from first sub-query if query_plan.sub_queries: self.primary_subintent = query_plan.sub_queries[0].subtype self.presentation_hint = query_plan.presentation_hint.value if query_plan.presentation_hint else "text" # Set classification metadata if classification_result: self.llm_model_used = classification_result.llm_model self.tokens_used = classification_result.tokens_used self.classification_time_ms = classification_result.classification_time_ms self.classification_reasoning = classification_result.reasoning def get_sub_query_count(self) -> int: """Get number of sub-queries""" return len(self.sub_queries) if self.sub_queries else 0 def get_all_intents(self) -> List[str]: """Get list of all intents in this query""" if not self.sub_queries: return [] return [sq.get("intent") for sq in self.sub_queries if sq.get("intent")] def get_all_subintents(self) -> List[str]: """Get list of all sub-intents in this query""" if not self.sub_queries: return [] return [sq.get("subtype") for sq in self.sub_queries if sq.get("subtype")] def has_intent(self, intent: str) -> bool: """Check if query contains a specific intent""" return intent in self.get_all_intents() def has_subintent(self, subintent: str) -> bool: """Check if query contains a specific sub-intent""" return subintent in self.get_all_subintents() def get_parallel_sub_queries(self) -> List[Dict]: """Get sub-queries that can be executed in parallel""" if not self.is_multi_intent or "parallel" not in (self.intent_relationships or []): return [] return self.sub_queries or [] def get_sequential_sub_queries(self) -> List[Dict]: """Get sub-queries that must be executed sequentially""" if not self.is_multi_intent or "sequential" not in (self.intent_relationships or []): return [] return self.sub_queries or [] def update_with_enhanced_result(self, enhanced_result): """Update query with enhanced result data""" from app.schemas.intent import EnhancedQueryResult if isinstance(enhanced_result, EnhancedQueryResult): self.result_summary = enhanced_result.combined_result self.result_explanation = enhanced_result.overall_explanation self.confidence_score = enhanced_result.overall_confidence self.result_provenance = enhanced_result.all_provenance self.data_quality_notes = enhanced_result.data_quality_notes self.suggested_actions = enhanced_result.next_steps self.related_queries = enhanced_result.related_queries self.elements_processed = enhanced_result.total_elements_processed self.total_processing_time_ms = enhanced_result.total_processing_time_ms self.cache_hit = enhanced_result.cache_hit # Set result count from primary result if enhanced_result.sub_results: self.result_count = enhanced_result.primary_result.elements_processed def to_dict_enhanced(self) -> Dict[str, Any]: """Convert to dictionary with enhanced multi-intent fields""" base_dict = self.to_dict() base_dict.update({ "is_multi_intent": self.is_multi_intent, "sub_queries": self.sub_queries, "intent_relationships": self.intent_relationships, "primary_intent": self.primary_intent, "primary_subintent": self.primary_subintent, "presentation_hint": self.presentation_hint, "llm_model_used": self.llm_model_used, "tokens_used": self.tokens_used, "classification_time_ms": self.classification_time_ms, "classification_reasoning": self.classification_reasoning, "result_provenance": self.result_provenance, "data_quality_notes": self.data_quality_notes, "suggested_actions": self.suggested_actions, "related_queries": self.related_queries, "elements_processed": self.elements_processed, "total_processing_time_ms": self.total_processing_time_ms, "plain_text_prompt": self.plain_text_prompt, }) return base_dict class SubQueryResult(Base): """ Individual sub-query results for multi-intent queries """ __tablename__ = "sub_query_results" id = Column(Integer, primary_key=True, index=True) query_id = Column(Integer, ForeignKey("queries.id", ondelete="CASCADE"), nullable=False, index=True) sub_query_index = Column(Integer, nullable=False) # Sub-query identification intent = Column(String(50), nullable=False, index=True) subtype = Column(String(100), nullable=False, index=True) query_fragment = Column(Text, nullable=False) confidence = Column(Float, nullable=True, index=True) # Results result = Column(Text, nullable=True) result_data = Column(JSON, nullable=True) explanation = Column(Text, nullable=True) provenance = Column(JSON, nullable=True) # Element IDs processing_steps = Column(JSON, nullable=True) limitations = Column(JSON, nullable=True) # Performance metrics elements_processed = Column(Integer, nullable=True) processing_time_ms = Column(Integer, nullable=True) created_at = Column(DateTime(timezone=True), server_default=func.now()) # Relationship query = relationship("Query", back_populates="sub_query_results") def __repr__(self): return f"" def to_dict(self) -> Dict[str, Any]: """Convert to dictionary for API responses""" return { "id": self.id, "sub_query_index": self.sub_query_index, "intent": self.intent, "subtype": self.subtype, "query_fragment": self.query_fragment, "confidence": self.confidence, "result": self.result, "result_data": self.result_data, "explanation": self.explanation, "provenance": self.provenance or [], "processing_steps": self.processing_steps or [], "limitations": self.limitations or [], "elements_processed": self.elements_processed, "processing_time_ms": self.processing_time_ms, "created_at": self.created_at.isoformat() if self.created_at else None, } @classmethod def create_from_single_result(cls, query_id: int, index: int, single_result): """Create SubQueryResult from SingleQueryResult""" from app.schemas.intent import SingleQueryResult if isinstance(single_result, SingleQueryResult): return cls( query_id=query_id, sub_query_index=index, intent=single_result.intent.value, subtype=single_result.subtype, query_fragment=single_result.query_fragment, confidence=single_result.confidence, result=single_result.result, result_data=single_result.result_data, explanation=single_result.explanation, provenance=single_result.provenance, processing_steps=single_result.processing_steps, limitations=single_result.limitations, elements_processed=single_result.elements_processed, processing_time_ms=single_result.processing_time_ms, ) class QueryResultStorage(Base): """ Separate table for storing large query results """ __tablename__ = "query_result_storage" id = Column(Integer, primary_key=True, index=True) query_id = Column(Integer, ForeignKey("queries.id", ondelete="CASCADE"), nullable=False, unique=True) # Large result data stored separately full_result_data = Column(JSON, nullable=True) # Complete results raw_sql_result = Column(JSON, nullable=True) # Raw SQL results processed_elements = Column(JSON, nullable=True) # Processed element data # File-based storage for very large results result_file_path = Column(String(500), nullable=True) # S3 path for large results result_file_format = Column(String(20), nullable=True) # json, csv, excel # Result metadata result_size_bytes = Column(Integer, nullable=True) compression_used = Column(Boolean, default=False) created_at = Column(DateTime(timezone=True), server_default=func.now()) # Relationship query = relationship("Query", backref="query_result")