Afetharita disaster map uses easyOCR, Turkish BERT models, and Hugging Face Inference API on Spaces

AI Impact Summary

Afetharita is a disaster-response workflow that converts survivor-sourced data and social content into structured, geolocated needs. It combines OCR (easyOCR) to extract text from images, Turkish NLP models (e.g., bert-base-turkish-uncased, bert-base-turkish-cased) and NLI variants (xlm-roberta-large-xnli, convbert-base-turkish-mc4-cased-allnli_tr) for labeling, and a geocoding API to map coordinates on a front-end map, with deployment and scaling via Hugging Face Spaces and Inference API. The stack also leverages Gradio, Argilla for labeling/evaluation, and external imagery from Planet Labs/Maxar for context, enabling rapid end-to-end data-to-action in the field. This capability accelerates rescue coordination by turning multi-channel survivor signals into actionable heatmaps, but relies on external ML services and requires ongoing data drift management to maintain accuracy.