Featured Project
VisionPlays — CV + LLM Game Agent
Computer Vision · LLM Reasoning · Real-Time Game Play
An AI agent that plays TicTacToe by combining computer vision to read the physical board state with an LLM to reason about the optimal next move. The vision pipeline detects the grid, classifies X/O marks, and feeds the board state to the language model, which returns its move in real time.
Demo
How It Works
1. Board Detection (CV): Computer vision captures the live board, detects the 3×3 grid, and classifies each cell as empty, X, or O.
2. State Encoding: The detected board state is serialized into a structured representation the LLM can understand.
3. LLM Reasoning: The language model receives the current board state, evaluates possible moves, and returns the optimal next move with its reasoning.
4. Move Execution: The agent's chosen move is displayed in real time, completing the perception-reasoning-action loop.