* Automated Chess Commentary Generator

Summary

Spring/Summer of 2024, chess was blowing up on social media. I wanted to create a chess commeentary generator that would be able to generate qualitative, insightful, context-aware commentary. I actually started this project with a custom encoder-decoder architecture. However, I was not able to generate insightful commentary due to the limited amount of data I had. I then pivoted to using Llama-3 since it just came out, and fine-tuning it with a specially curated dataset to better capture the nuances of chess.

Brief details

• Applied innovative approach to chess commentary generation by fine-tuning Llama-3 with specially curated dataset
• Produced insightful, context-aware commentary through advanced prompt engineering with Llama-3 by utilizing FEN/UCI
• Leveraged LLM teacher-student framework to evaluate model performance and generalization