Backgammon Game Validation
Automated verification and formatting of backgammon game moves into standardized XG files.
Client:
Backgammon competition company, Armenia
Project Overview
The client engaged Axis to develop a system that validates game histories of backgammon matches, captured via computer vision, and converts them into the XG (eXtreme Gammon) format. The solution ensures that every detected move is legal according to the rules of backgammon, enabling players and analysts to archive and replay matches in professional tools.
Challenge
Initial client expectation was to validate move sequences with an LLM (e.g., Qwen).
High risk of “hallucinations” and unreliable logic when relying solely on generative AI for strict rule enforcement.
Extremely tight deadlines (1 week), with the requirement to deliver a functional validation tool before a live event.
Tech Stack
Computer Vision Input: YOLO – existing system for detecting checker positions and move sequences from video.
Validation Engine: Deterministic backgammon rules algorithm developed to step through and check the legality of every move.
File Format: XG (eXtreme Gammon) – industry-standard format for storing and sharing games.
AI Models (Fallback): Gemini (via GCP) proposed as an alternative validation and classification tool when strict algorithmic logic fails.
Infrastructure: Google Cloud Platform for scalable API calls and LLM processing.
Solution
Axis designed a two-path architecture:
Primary Path – Algorithmic Validation: A rules-based simulator was implemented to replay each detected move, rejecting any illegal actions.
Fallback Path – LLM Support: If the algorithm output mismatches expectations, Gemini is used to refine or confirm move sequences, ensuring deadlines are met.
Output: Verified games are formatted into
.xgfiles with match details.
