🔍 Coercive-Control Chat Analyzer

Fine-tuned Qwen3-14B that screens WhatsApp conversations for indicators of coercive control — classifying every message in its conversational context (up to 7 preceding messages) against a 7-category taxonomy grounded in domestic-violence research.

Try it: click the example file below, then 1 · Load chat → 2 · Analyze.

⚠️ Research prototype — please read. This is a screening aid, not a determination of abuse; false positives and false negatives occur. Do not upload real private conversations — this is a public demo running on shared infrastructure, and chat content is processed on Hugging Face servers. Use the synthetic example, or your own fictional test chats.

Model: Qwen3-14B + fine-tuned adapters · 4-bit NF4 (as trained and evaluated)