LAMDASZ-ML/ChinaTravel

Combines symbolic constraint verification with neural language models through a hybrid LLM-NeSy agent architecture, enabling complex itinerary planning under hard constraints (budget, transportation, timing). Includes a comprehensive Chinese travel database with annotated constraint DSLs, local LLM support (Qwen, Llama, Mistral), and an LLM-modulo baseline for iterative refinement using ground-truth symbolic validators. Extensible framework allowing custom agent development via BaseAgent inheritance and support for both API-based (DeepSeek, GPT-4o) and local model inference.