Klang Valley Flood Watch

Klang Valley Flood Watch is a situational-awareness map for pluvial flash-flood risk in Malaysia's Klang Valley, built entirely on free public data — no proprietary feeds, no paid APIs.

On load, it geolocates the viewer and asks Cloudflare Workers AI to compose a short, plain-language advisory for that exact spot, built only from facts it actually has: the nearest DID/JPS river-level station, active MetMalaysia thunderstorm warnings, any nearby documented past flash-flood case, Open-Meteo's short-range rain/CAPE/pressure signals, and a caution if a station upstream on the same named river is already elevated. The model is instructed never to invent a number or warning, and to state its own limits plainly — the upstream caution is explicitly labelled a name-match-plus-elevation heuristic, not a verified flow-network model, both in the UI and in the prompt itself.

Resilience is the core design constraint, not an afterthought. If the DID/JPS feed goes stale, the status pill switches to an explicit "Stale since HH:MM" state instead of silently serving an old reading as current. If the AI call itself fails, a deterministic, template-built advisory using the same underlying facts takes over so the panel never goes blank. Because real conditions are calm most of the time, the site includes six on-demand simulated scenarios (normal, worsening water with an active warning, a documented flood case, contradictory upstream signals, missing data, and a stale feed) so a reviewer can see all of this resilience behavior immediately, without waiting for an actual event — each one flies to its own simulated location, drops clearly-labelled "SIMULATED" pins for the station or case it's citing, and frames them all in view automatically.

Coverage is intentionally scoped to the Klang Valley, stated as such in the UI, and the project is explicit everywhere that it is not an official warning system.