sovanmondal/tayari

# 🌍 Tayari ### The Impact-Based Forecasting Co-pilot — turning early *warning* into early *action* *From ICPAC's forecast → to a life-saving decision → to the last mile.* `FastAPI` · `PostGIS` · `geopandas` · `React` · `MapLibre` · `Groq (Llama-3.3)` · `Docker` **100% real open data · 29 automated tests · one-command deploy** ### 🎥 Watch the demo github.com ▶️ Full quality on YouTube: **youtu.be --- ## ⚡ The 30-second story > Between **2020 and 2023, the Horn of Africa suffered its worst drought in four decades.** > More than **20 million people** faced crisis-level hunger. Millions of livestock — the entire > savings and livelihood of pastoralist families — perished. > > Here is the uncomfortable part: **the forecasts were right.** ICPAC saw it coming months ahead. > The warnings existed. And still, the response arrived too late. Why? Because **seeing a drought and acting on it are two different problems.** - ICPAC already produces world-class forecasts. - HUSIKA already delivers alerts to the last mile. - But the step *in between* — deciding **who exactly is at risk, what action to trigger, and why** — is still done by hand, district by district, in spreadsheets and PDFs. **It doesn't scale.** ICPAC calls this the *"long-pending operationalization of Impact-Based Forecasting."* It is the **last-mile gap** — and it is where lives and livelihoods are lost. **Tayari is the missing middle.** It plugs into ICPAC's real data and, for every district, answers the three questions that turn a forecast into action — with an auditable trail — then writes the message that reaches the herder, in their language, on their basic phone. > **Tayari doesn't compete with ICPAC's tools. It completes the stack.** --- ## 🧭 Table of contents - Where Tayari fits - Architecture - How the reasoning works - Features - Real data sources - Responsible AI - Tech stack - Run it locally - A …