Notes on grounding a travel planner in verifiable data — where the facts come from, where they stop, and the decisions behind the things that look like small print.
13 articles
2 min read
Armenia offers ancient monasteries, dramatic mountain scenery, welcoming cities and excellent local food—all within a relatively compact country. However,
2 min read
Weather, price and crowds do not peak in the same week. A month name is not an answer — it is where the real question starts.
2 min read
You land tired, at an hour you did not choose, in a city whose transport you have never used. That is the day a plan has to earn its place.
4 min read
Millions of people hold a residence permit that opens doors their passport does not. For a year our entry rules ignored every one of them.
3 min read
If you do not name a month, we do not quietly use this one. Here is the arithmetic that picks a month instead, and how to tell when it did.
4 min read
Between the model finishing a section and that section reaching your screen, ten code-only checks run. This is all ten, and what each one caught.
3 min read
Language models are fluent about places that do not exist. Here is the architecture we use to make sure ours cannot be.
2 min read
We show visa status for your passport on every plan. This is an honest account of where that data comes from and where it stops.
2 min read
Every figure in a Saferni budget carries the same label. That is a deliberate design decision, not legal throat-clearing.
2 min read
“Kerala” is not a place you can book a hotel in. Turning a region, an island or a whole country into somewhere you can actually sleep is its own problem.
3 min read
Asking a model for sixty days in one go does not produce a long plan. It produces no plan at all. Here is what we do instead.
3 min read
If the entities were never the model's to invent, it turns out the sentences around them need not be either. We built the whole plan a second time, in code.