What To Do Today — three concrete things, in the next four hours

Why this exists

The most expensive Saturday is the one where you sit at home, agree with yourself that you should do something, and then doom-scroll until 7pm. The cost is rarely the activity — it's the decision. Picking what to do is harder than doing it.

Most "what to do near you" apps are tourism-shaped. They suggest restaurants. They surface popular places. They don't know it's 2pm, drizzling, and you've got ₹500 in your pocket and three hours before dinner.

What it does

Three inputs:

1. City (or auto-detect)
2. Budget for the afternoon — slider, ₹0 to ₹2,000
3. Mood — three buttons: active, quiet, social

The app checks weather via Open-Meteo (free, no API key needed). Calls Gemini with the inputs and gets three concrete suggestions — go walk Lodhi Gardens for an hour, bookshop crawl around Khan Market, catch the 3:45 show at PVR Saket. With approximate cost and travel time per suggestion. No ranked list of fifty options. Three. The whole point is that picking three already happened for you.

A "next" button cycles to three more if none land.

How it's built

• Frontend: a single page. Form → results.
• API: Open-Meteo for weather (no key), Gemini for suggestions (cheap).
• No accounts. No history. The app is stateless on purpose — every session is a fresh decision.

What it isn't

Not a discovery platform. Not a booking system. The suggestions are prompts, not links — the user goes and does, the app doesn't try to insert itself between intent and action.

Not a tourism app. The user is a local who's bored, not a visitor.

Why it doesn't exist yet

The hard problem is the suggestion quality, not the engineering. Generic Gemini suggestions read like a travel blog from 2018. The prompt needs city-specific knowledge — Bombay's Bandra is different from Bombay's Andheri — and that needs either a manual taste pass per city or grounded search results that bias to local sources.

I'm parking this until I figure out which approach scales.