muriithi-alvin/Husika-Action_Pipeline

Husika is an automated, event-driven data pipeline that bridges the gap between complex meteorological data and last-mile communities. It continuously monitors hazard thresholds. When a threshold is breached, it triggers a Generative AI layer that translates the raw data into actionable SMS alert in local languages like Swahili or Somali. # 🌍 Husika-Action Pipeline > Transforming ICPAC early warning data into actionable, last-mile SMS alerts using GenAI and MLOps principles. ## 🚀 Project Overview Rural communities in the IGAD region face a critical "last-mile" gap in early warning systems. Hazard bulletins are often technical, lengthy, and in English, while target communities may have low network coverage and speak local languages (e.g., Swahili, Somali). **Husika-Action Pipeline** bridges this gap. It is an automated, event-driven data pipeline that monitors hazard thresholds, uses a quantized LLM to generate hyper-local, plain-language SMS alerts (<160 chars), and dispatches them via low-bandwidth gateways. ## 🏗️ Architecture & Tech Stack - **Data Ingestion & Rules Engine:** Python, Custom Threshold Logic - **GenAI Layer:** Groq API (Llama-3-8b-instant) for low-latency, constrained summarization and translation - **Last-Mile Delivery:** SMS Gateway (with robust Mock/Sandbox fallback for resilience) - **MLOps & Deployment:** Docker, GitHub Actions (CI/CD), Structured Logging ## 🛠️ Local Development & Execution ### Prerequisites - Python 3.11+ - Docker (optional, for containerized run) ### Setup 1. Clone the repository: ```bash git clone github.com cd husika-action-pipeline