Core Processing

The core processing component contains the central processing logic and model implementation for the Four-Sided Triangle application.

Components

Model Implementation (model.py)

The SprintLLM class handles the core model functionality:

• Loading and management of the domain expert LLM
• Query processing and response generation
• Anthropometric metrics calculation
• Utility functions for prompt formatting and response extraction

Key functions:

• load_model(): Loads the sprint-llm-distilled model and tokenizer
• generate_response(): Generates responses to user queries
• calculate_anthropometric_metrics(): Calculates metrics based on user parameters
• get_model_instance(): Factory function to get a model instance

Modeler (modeler.py)

The Modeler class acts as a bridge between query processing and solving:

• Transforms unstructured queries into structured entity-relationship models
• Extracts entities, relationships, parameters, and constraints
• Enriches models with domain knowledge
• Integrates components into unified knowledge models

Key functions:

• process_query(): Processes query packages into knowledge models
• _extract_model_components(): Extracts structured data from queries
• _enrich_with_domain_knowledge(): Adds domain-specific context
• _integrate_model_components(): Creates unified knowledge models
• get_modeler_instance(): Factory function to get a modeler instance

Processing Pipeline

The core processing is organized into distinct stages, each handling a specific aspect of query processing:

Stage 0: Query Processing

• Initial query analysis
• Intent classification
• Parameter extraction
• Query validation

Stage 1: Semantic ATDB

• Semantic analysis
• Throttle detection
• Behavioral pattern recognition
• Performance optimization

Each stage is designed to be modular and maintainable, allowing for easy updates and improvements to individual components without affecting the entire pipeline.