I wanted to start a broader mistakes and lessons thread around Full-Stack AI Development.
A useful angle is data quality, but it also connects with real-world use. Different tools and workflows can produce very different results even when the goal is similar.
It would be interesting to compare...