A prototype only needs to work once. A production system has to keep working when the prompt changes, the model drifts, a dependency slows down, or the input stops looking like the test set.
That gap is where most of the useful engineering happens: evaluation, observability, failure handling, and deciding when the model should not answer at all.
This notebook is a place for those details—the experiments that failed, the trade-offs that survived, and the patterns worth carrying into the next build.