CSE 291A Winter 2025 Project

Exploring LLM-augmented game-playing agents across multiple domains.

Repository Structure

Directory	Description
Tool_Creation/	MCTS + LLM Heuristic Optimization — a modular Monte Carlo Tree Search framework with pluggable phase tools and an automated LLM-driven improvement pipeline. See Tool_Creation/README.md for full documentation.
connect_four/	Connect Four MCTS experiments (Jupyter notebook)
tic_tac_toe/	Tic-Tac-Toe MCTS experiments (Jupyter notebook)
quoridor/	Quoridor AI with C++ MCTS engine and GUI
games/	Standalone game implementations (Uno)
play_zork/	LLM-powered Zork agent using vision-based screen reading

Quick Start — MCTS + LLM Pipeline

cd Tool_Creation
./run_pipeline.sh            # run the full optimization loop
./run_pipeline.sh --iters 10 # override iteration count

See Tool_Creation/README.md for detailed setup and usage.

Requirements

• Python 3.12+
• An OpenAI-compatible API key (set in Tool_Creation/.env) for the LLM pipeline
• pytest for running the test suite