QERRA-v2 Classical Edition – Full SEMEV-12 Implementation & Live Public API
Solo researcher releases transparent AI ethics layer with 12-vector scoring and live API.
Solo researcher Marussa Metocharaki has publicly released QERRA-v2 Classical Edition, a fully operational implementation of the SEMEV-12 ethical framework. The system now runs all 12 vectors with real weighted ethical scoring, going beyond simple rule-based checks. Four of the most semantically nuanced vectors—harm intent, cognitive manipulation, autonomy violation, and institutional trust—leverage sentence-transformers for context-dependent detection, allowing the system to understand subtle language cues that simpler classifiers would miss. The API also includes dynamic reasoning output, human-readable risk explanations, and proper multi-vector activation.
The public API is live and open for testing on Hugging Face Spaces at https://qerra-v2-api-classical-qerra-v2-api-classical.hf.space/docs. Metocharaki emphasizes this is the classical (non-quantum) foundation of a hybrid project, built step-by-step as a solo researcher. The goal is to create a transparent, human-centered ethical decision layer for humanoid robots and AI systems. The researcher is actively seeking feedback from the robotics and AI safety community, particularly on how this layer could integrate with ROS2-based systems for real-time ethical deliberation in autonomous agents.
- Full SEMEV-12 implementation with 12 active ethical vectors, each contributing a weighted score.
- Four vectors (harm intent, cognitive manipulation, autonomy violation, institutional trust) use semantic detection via sentence-transformers.
- Live public API on Hugging Face Spaces with dynamic reasoning and human-readable risk explanations.
Why It Matters
Ethical scoring layers like QERRA-v2 could become the transparency standard for autonomous robot decision-making in critical applications.