prompt-attack

Build production ready apps for GPT using Node.js & TypeScript

Adversarial attack and robustness evaluation for Chinese fraud text classification. Adapts PromptAttack (ICLR 2024) with LLM-generated adversarial paraphrases, SQLite API caching, and RoBERTa fine-tuning.

🔒 Safeguard LLM behavior with PromptGuard to detect unseen regressions and ensure reliable outputs amid evolving model updates.