Popular ML components like LSTMs, Transformers, CNN etc. Building ML models in the context of speech recognition/NLP and advantages or shortcomings of each approach. These have to all be addressed while building an ML pipeline. Know your own projects in detail and why certain models were used vs. others