Technical blog
Notes from building MooTrip, an AI trip planner: what the model does, what code does, and what went wrong on the way.
Model or code? The pipeline behind an AI trip planner, and which steps are deterministic
The steps one edit goes through in an AI trip planner, which of them the model does and which ordinary code does, and three steps that changed sides.
10 October 2026
Testing an AI feature: what my evals changed my mind about
Adding a rule to the prompt made edits worse. Six things a small test set taught me about building on a language model.
10 October 2026
Making an AI feature faster and cheaper: eight things that worked
One edit went from minutes to 14 seconds. Eight changes that made an AI feature faster or cheaper, each as a rule you can reuse.
10 October 2026
Monitoring an AI feature: what to record so you can see why it failed
Seven things to record so that when an AI feature fails, you can find out why.
10 October 2026
Refactoring an AI pipeline: what I cleaned up, and what it fixed
Seven changes that untangled the code around a language model: named steps, checks apart from repairs, one definition of the format.
10 October 2026
Two weeks building MooTrip: a retro on what I built
A first version built in three days and replaced on the fourth: what went well, what did not, and what I would do differently.
10 October 2026
Two weeks building MooTrip: a retro on how I worked
A sprint retro on two weeks of solo work, read back from the git history: pace, commit habits and designing by trial.
10 October 2026