Show HN: Ada – A governed AI that validates before generating (github.com)

Ada is a newly introduced governed AI conversational interface that prioritizes validation over mere generation. Unlike typical AI systems that produce responses based on prompts without a pre-validation step, Ada first analyzes the input through a layer called Newton. Newton rigorously validates prompts, allowing generation only if they meet specific structural criteria. This approach helps prevent harmful outputs, hallucinations, and logical contradictions, ensuring that users receive reliable responses while maintaining control over the conversation flow. The significance of Ada lies in its capacity to manage potentially adversarial inputs without relying on complex machine learning techniques. By employing a simple, deterministic pattern-matching mechanism, Ada successfully blocks 87.5% of risky prompts, enabling it to sidestep common pitfalls such as self-negation and false claims. Its design allows for human review of flagged queries, reducing the chances of misinformation while enhancing user trust. With no external dependencies and straightforward implementation in Swift, Ada represents a notable shift in AI conversational design, where integrity and user empowerment take precedence over engagement metrics.