afetharita disaster map uses easyOCR, Turkish NER models, and Hugging Face Inference API for survivor data extraction

AI Impact Summary

Volunteer-led afetharita translates survivor-reported text from tweets and screenshots into structured data (name, phone, address) using easyOCR for OCR, Turkish NER models, and a geocoding API to place responders on a map. The stack leverages Gradio and Hugging Face Spaces for rapid prototyping, with Hugging Face Inference API serving models at scale, and a CI-driven ephemeral environment to test changes. The team experimented with few-shot prompting and NLI models to adapt to data drift, maintaining a leaderboard and Argilla-based labeling to benchmark recall and F1. This open-source, cloud-assisted approach accelerates field readiness, but requires strong data governance and privacy safeguards due to handling personal information.