IBM/responsible-prompting-api

Flask-based API that provides prompt recommendations via semantic similarity matching using sentence transformers, comparing user inputs against curated datasets of responsible-practice prompts and flagged adversarial examples. Supports multiple model backends including HuggingFace and Replicate, with customizable thresholds for "add" and "remove" recommendations and built-in visualization through t-SNE embeddings. The `recommend` endpoint returns scored prompt suggestions aligned with social values like fairness, inclusion, and transparency, enabling developers to integrate responsible prompting into LLM workflows.