scvagritech/tawi-climate-smart-nutrition-intelligence-system

AI-powered produce quality assessment for school feeding programmes. Computer vision freshness grading + climate-linked spoilage prediction, built for edge deployment in low-connectivity environments. Open-source under MIT. Extracted from a live production system serving schools in Kenya. # Tawi Fresh Climate Smart Nutrition Intelligent System **Open-source, edge-deployable produce quality assessment for school feeding supply chains in climate-vulnerable regions.** --- ## The Problem Kenya's school feeding programmes serve millions of children daily, but the fresh produce supply chains behind them operate without any real-time intelligence on food quality, safety, or nutritional adequacy. Climate stress — rising temperatures, extended transit without cold chain, and erratic rainfall — has made quality degradation at delivery a near-daily operational reality. No scalable, data-driven system currently bridges climate exposure, produce quality, and child nutrition outcomes in school feeding supply chains. ## What This Project Will Contain Tawi Fresh Climate Smart Nutrition Intelligent System is the open-source inference and assessment module extracted from Tawi Fresh's production platform. It provides: | Component | Description | Status | |-----------|-------------|--------| | **Quality Grading Model** | TensorFlow/TFLite computer vision model for produce freshness classification across five quality indicators (size specification, pest infestation, rot, dehydration, discolouration) | 🔨 In development | | **Edge Inference Engine** | Offline-capable inference pipeline using OpenCV and TensorFlow Lite, designed for low-connectivity school environments | 🔨 In development | | **Climate-Quality Correlation Engine** | Rules-based engine linking quality scores to climate exposure variables (ambient temperature, transit duration, source-region weather) to generate predictive spoilage risk scores | 🔨 In development | | **Training Data Toolkit** | Labelling utilities, augmentation pipelines, and data preparation scripts for building produce quality datasets | 🔨 In development | | **Anonymised Training Dataset** | Labelled produce images across seasonal and climate variations, published as an open public good | 📋 Planned (Month 7–9) | ### Supported Cro …