FlyMeHere

The problem

Finding a cheap flight in Southeast Asia usually means wrestling with rigid search forms — and even then, you miss the small regional carriers that often have the best fares. The big engines don't cover the region well, and neither of them will quietly keep watching a route for you and speak up when the price moves.

How it worked

Instead of a form, you'd describe the trip — "Bali next month with my family on a budget" — and it would work out what you actually meant: where you're going, roughly when, how flexible you are, who's travelling and what you want to spend. That part got built, and it worked. Claude handled the intent extraction, and the plan from there was to feed those parameters into a flight pricing API, rank what came back, and keep watching the route in the background.

Why I scrapped it

There was no flight pricing API I could actually buy. Every viable provider gates access behind something a pre-launch solo founder doesn't have — traffic you're expected to already command, or a jurisdiction and a legal entity you either have or you don't.

• Skyscanner — around 100K+ monthly active users, 500+ followers, and a content history under a year old.
• Kiwi — formerly the accessible option; the partner programme now expects roughly 50K+ monthly active users.
• Duffel — the best developer experience in the category, but live fares need KYC and business registration, and Singapore isn't a supported jurisdiction.
• Amadeus — the Self-Service tier, the one genuinely open door in the industry, was decommissioned on 17 July 2026.

What I took from it

I tried routing around it — a Telegram bot for distribution, a scraper-backed free tier for data — and neither survived contact with the arithmetic. The real diagnosis was that the problem had never been distribution or technology. The data itself was gatekept, and that had been the binding constraint the whole time.

So I scrapped it in July 2026, with the language layer working and nothing real to put behind it. Short and worth it: the wall announced itself as soon as I went looking, which is the whole skill — telling a structural problem apart from an execution problem before you've spent six months on it.

Stack