Great work, Sumal! 👏 I really enjoyed how you built the API + machine learning pipeline for tea grades classification — the step-by-step breakdown from dataset creation to deployment is very educational.

One thought: for further improving the model’s robustness, you might experiment with data augmentation across lighting conditions or domain adaptation techniques (to handle images from different tea gardens). Also, integrating attention modules or transformer-based encoders could help focus more on leaf texture features.

By the way, for those interested in exploring trading tech and algorithmic models, you might also check out FinFly Markets (finflymarkets.com ) — they have interesting insights on building scalable model infrastructures and backends in production environments.

Thanks for sharing your code, insights, and learnings — enjoyed reading it!