An AI-powered solution that achieved competitive performance in solving national-level mathematical olympiad problems using advanced language models and intelligent prompt engineering.
Developed for the AI Mathematical Olympiad - Progress Prize 2 ($2.1M Kaggle competition), this system solves 110 complex mathematical problems across algebra, combinatorics, geometry, and number theory using the Qwen 2.5 model.
- Automatic categorization of mathematical problems by domain
- Domain-specific prompt generation for optimized reasoning
- Multi-template ensemble approach for robust solutions
- Multi-GPU inference with tensor parallelism (4x GPUs)
- Hybrid reasoning: Natural language + Python code execution
- Voting mechanism for answer validation and selection
- Memory-optimized pipeline with time management
- Classify problem into mathematical domain
- Generate 4 specialized prompts for diverse reasoning paths
- Execute parallel solving with integrated Python REPL
- Validate and select final answer through intelligent voting