🏆 Proud to share that our team ranked 33rd out of 2,674 teams in Kaggle's Deep Past Challenge – Translate Akkadian to English, earning a Silver Medal.
This challenge focused on reviving 4,000-year-old Assyrian texts using AI and machine translation.
🔹 Fine-tuned ByT5-base with a two-stage training strategy
🔹 Trained over 21 epochs through iterative experimentation
🔹 Used only public data — no synthetic data, no LLM-generated pairs, no ensembles
🔹 Achieved a private score of 38.1 and secured a Top 1.3% finish
What I enjoyed most wasn't the final ranking, but the process: testing ideas, learning from failures, and proving that disciplined experimentation and data quality can compete with far larger resources.
Research, NLP, LLMs, and low-resource language problems continue to fascinate me. Excited to apply these lessons to future AI systems and real-world applications.
#Kaggle #MachineLearning #DeepLearning #NLP #LLM #AI #DataScience #Research #Python #ByT5 #TranslationAI #MLOps