dcai-course/dcai-lab

Covers nine progressive labs spanning label error detection via Confident Learning, multi-annotator dataset curation, outlier identification, active learning, feature interpretability, and membership inference attacks. Jupyter notebooks combine hands-on implementation of data-centric techniques (data quality improvement, augmentation, prompt engineering) with black-box model evaluation, emphasizing how dataset engineering outperforms model-centric optimization. Integrates with standard ML libraries and LLMs, with Labs 8+ supporting Colab execution for zero-setup experimentation.

479 stars. No commits in the last 6 months.