Take a look at https://docs.inkeep.com/ai-api/openai-chat-completion-endpoint#question-answer-mode https://docs.inkeep.com/ai-api/openai-chat-completion-endpoi.

Take a look at https://docs.inkeep.com/ai-api/openai-chat-completion-endpoint#question-answer-mode https://docs.inkeep.com/ai-api/openai-chat-completion-endpoi... tl;dr we define a JSON schema with a few semantic labels that represent a gradient in confidence. On our end we prompt it with certain examples and guidance for when to use each label. This is generally a better approach than e.g. asking an LLM to give a numeric score. We also have trained embedding-based classifiers as non-LLM heuristics.