Your company manages an ecommerce platform and has a large dataset of customer reviews.

The keywords here are " quickly prototype " and " minimize time and cost. "

Natural Language API: Google’s pre-trained Natural Language API provides out-of-the-box sentiment analysis. It requires zero training time, zero infrastructure management, and no labeling effort. This is the fastest and most cost-effective way to get a baseline/prototype.

Why other options are incorrect:

Option A: Fine-tuning a BERT model is computationally expensive and requires significant engineering time.

Option C: AutoML is a great secondary step if the API is not accurate enough, but it still requires time to train and evaluate, making it slower than the pre-trained API.

Option D: Vectorizing text and training a regression model is unnecessarily complex for a standard sentiment classification task.