Zimbabwe has 90+ towns and cities, a population of 15+ million, significant agricultural and mining sectors — and almost no weather infrastructure built specifically for it. Global apps cover Harare and Bulawayo at best, and the AI summaries they generate read like they were written for someone in London. I built mukoko weather to fix this. A few things that shaped the approach: Weather as a public good. The platform is free, no ads, no paywalls. If a smallholder farmer in Chinhoyi needs frost risk data to protect their crops tonight, that can’t be behind a subscription. Hyperlocal context matters more than raw data. Zimbabwe has distinct agricultural seasons — Zhizha (rainy), Chirimo (spring), Masika (early rains), Munakamwe (winter). Elevation varies dramatically: the Highveld sits above 1,200m, the Zambezi Valley below 500m. The AI assistant, Shamwari Weather, is prompted with this geographic and seasonal context so its advice is actually meaningful to the user. Constrained environments are the primary target, not an edge case. Mobile-first, bandwidth-efficient, PWA-installable. The majority of users are on Android, often on 3G or spotty LTE. That’s not a future concern — it’s the baseline. Technical decisions: ∙ Next.js 15 App Router on Cloudflare Pages + Workers ∙ AI summaries via Anthropic Claude SDK, server-side only, cached at the edge with KV and immutable TTL tiers (AI-generated weather advice shouldn’t change retroactively) ∙ Open-Meteo for forecast data (free, high-quality global model coverage) ∙ 90+ Zimbabwe locations validated against geographic bounds with elevation data for frost modelling The broader vision is weather as infrastructure within a larger Africa super app (Mukoko Africa), with Zimbabwe as the proof of concept before expanding to other developing country markets using the same locally-grounded approach. Would love feedback on the approach, especially from anyone who’s built for similar markets — low-bandwidth, mobile-dominant, regions underserved by global platforms.