Question: Explain why a team should avoid trying to design a perfect machine learning system before any prototype exists, especially when entering a new domain. In your answer, discuss limited domain knowledge, the difficulty of predicting the best technical direction in advance, and how error analysis helps guide improvement.

Sample answer: Trying to build a perfect system before having any results is usually a poor strategy. At the beginning of a project, the team often lacks enough domain knowledge to know which design choices will matter most. It is easy to spend a long time refining ideas that later turn out not to be the main bottleneck.

A better approach is to create a simple working prototype quickly, even if it is only good enough to establish a baseline. For example, a team building a classifier for sorting customer support tickets might first train a straightforward model in a few days rather than spending weeks on a complicated architecture. Once that baseline exists, the team can examine its mistakes.

Error analysis turns those mistakes into useful guidance. By reviewing where the prototype fails, the team can see patterns, prioritize the most important fixes, and choose the next experiments more intelligently. This iterative process is usually faster and more effective than trying to predict the final solution from the start.

Key points:

• New domains usually provide too little information to predict the best design choices in advance
• Trying to perfect the system first can waste time on the wrong ideas
• A simple baseline can often be built quickly
• Error analysis on the baseline reveals the most promising directions for iteration

Rubric: The response should address: 1) why limited domain knowledge makes it hard to choose the right design up front, 2) why a complex first attempt can waste effort, and 3) how a quick baseline plus error analysis supports faster, better iteration.