Show HN: Jevper – the Jev interface on top of any OpenAI-compatible model (github.com)

The recent launch of Jevper introduces a novel interface that allows users to seamlessly layer a new querying system—known as the Jev interface—over any OpenAI-compatible language model, such as a hosted LLM or a self-hosted llama.cpp server. This independent implementation enables developers to type questions and receive structured responses that include probabilities and confidence scores, all while maintaining compatibility without needing to modify existing code. Jevper's functionality does not rely on the TypeSafe API, allowing for greater flexibility in usage and deployment. This development is significant for the AI/ML community because it enhances the accessibility of advanced language model capabilities by standardizing the question-answering process across various backend systems. The technical framework supports three types of queries—Noul (binary yes/no), Choice (multiple labeled options), and Score (ranked scale)—allowing for diverse applications in fields like customer service, sentiment analysis, and billing inquiries. Moreover, Jevper leverages concurrency in its processing of questions, which can improve efficiency in real-time applications. As a result, this tool not only democratizes high-level ML interactions but also sets a foundation for more robust, varied AI use cases.